Sales Quantity Prediction System, Sales Quantity Prediction Method

Abstract

To provide a sales amount prediction system and a sales amount prediction method capable of predicting a further appropriate sales amount by comprehensively grasping a relation between data of a plurality of stores.SOLUTION: A sales amount prediction system includes a processor.The processor predicts an amount of sale of a commodity of an object store, by using a learned model which predicts an amount of sale of a commodity generated on the basis of data regarding the amount of sale of a commodity of each store in which handled commodities can be partly different between the stores .SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a sales volume prediction system and a sales volume prediction method.

Background Art

[0002] Conventionally, a technique for predicting the sales volume of a product using an appropriate model has been known. Patent Document 1 discloses a technique for performing prediction using data of a plurality of stores including data of other stores when predicting the sales of a target store.

[0003] That is, Patent Document 1 uses a plurality of past customer arrival number data as teacher data (training data), performs multiple regression analysis, and calculates the regression constant a, regression coefficients b1, and b2 of a calculation formula for predicting the customer arrival number of the target store (a calculation formula with y as the objective variable and x as the explanatory variable). Then, it multiplies the predicted customer arrival number value predicted by this calculation formula by the average transaction price to calculate the predicted sales value. Further, Patent Document 1 discloses that the prediction calculation may also be performed by applying, alone or in combination, methods such as random forest, neural network, generalized linear model, generalized additive model, ensemble learning model, and SVC, in addition to multiple regression analysis.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] For example, when there is data on multiple stores such as chain stores, operators and the like consider sales quantity prediction by taking into account differences in products handled and sales performance at each store (that is, based on the relationships between data of multiple stores grasped from an overall perspective). However, it is considered that in existing prediction systems, the relationships between data of multiple stores are not fully utilized for prediction.

[0006] Therefore, an object of the present invention is to provide a sales quantity prediction system and a sales quantity prediction method that can comprehensively grasp the relationships between data of multiple stores and perform more appropriate prediction of the sales quantity of products.

Means for Solving the Problems

[0007] According to a first aspect of the present invention, the following sales quantity prediction system is provided. That is, the sales quantity prediction system includes a processor. The processor predicts the sales quantity of products in a target store using a learned model that predicts the sales quantity of products generated based on data on the sales quantity of products in each store where the products handled between stores may be partially different.

[0008] According to a second aspect of the present invention, the following sales quantity prediction method is provided. That is, the sales quantity prediction method is a sales quantity prediction method performed using a processor. And this sales quantity prediction method (1) generates a learned model that predicts the sales quantity of products based on data on the sales quantity of products in each store where the products handled between stores may be partially different, and (2) uses the learned model to predict the sales quantity of products in a target store.

Effects of the Invention

[0009] According to the present invention, a sales quantity prediction system and a sales quantity prediction method that can comprehensively grasp the relationships between data of multiple stores and perform more appropriate prediction of the sales quantity of products are provided. Note that problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Embodiments for Carrying Out the Invention

[0011] With reference to the drawings, the sales volume prediction system will be described. The sales volume prediction system can predict the sales volume of a product (i.e., the sales quantity) of a target store using a trained model for predicting the sales volume of the product.

[0012] In this embodiment, in a situation where there are partial differences in the products handled by a plurality of stores forming a chain store, the task is to measure the sales quantity of each product in each store and predict the sales quantity for the next day based on the data up to the day when the measurement was made. The configuration and processing of a sales quantity prediction system that executes this task will be described. However, the present invention is not limited to this example. The sales quantity prediction system may, for example, predict the sales quantity based on the measured data of the sales quantity in a situation where there are no partial differences in the products handled by a plurality of stores. Further, the sales quantity prediction system may predict the sales quantity based on the data acquired over a predetermined period. Also, the sales quantity prediction system is not limited to predicting the sales quantity for the next day and may be used, for example, to predict the sales quantity for a certain time period.

[0013] As shown in FIG. 1, the sales quantity prediction system 1 includes a sales quantity measurement unit 101 and a controller unit 102. The controller unit 102 further includes a measurement data collection unit 103, a measurement data interpolation unit 104, a statistic calculation unit 105, a correlation matrix calculation unit 106, a correlation coefficient correction unit 107, a graph generation unit 108, an attribute data addition unit 109, a temporal feature processing unit 110, a graph feature processing unit 111, a feature integration processing unit 112, and a prediction result output unit 113. Here, the above-described various configurations (103 to 113) are programs and are stored in an appropriate storage device. In the controller unit 102, a deep learning unit 114 that represents functions related to the execution of deep learning in the learning phase is configured by programs (110, 111, 112).

[0014] The sales quantity measurement unit 101 is responsible for processing to measure, for each product sold, data such as product classification (such as detergents and snacks) and the model number unique to each individual product (assuming the product name is also unique, it is regarded as the model number), the sales quantity, and the date and time of sale, in cooperation with a POS (Point of Sale) system installed in each store. Then, the data measured by the sales quantity measurement unit 101 is transmitted to the controller unit 102.

[0015] As an example, the controller unit 102 is arranged at the headquarters of a chain store (i.e., the center that manages all of the multiple stores). The controller unit 102 is responsible for collecting the data measured by the sales quantity measurement unit 101 from each store, performing learning on the prediction of the sales quantity based on this data, and performing processing for predicting the sales quantity based on the learning results.

[0016] Each program (103 to 113) will be described. Here, the processor of the controller unit 102 is the main body of the processing of each program (103 to 113). The measurement data collection unit 103 is responsible for the process of collecting the data measured by the sales quantity measurement unit 101 via the Internet or the like. The measurement data interpolation unit 104 is responsible for the process of interpolating the time-series data of the sales quantity of products not commonly handled in multiple stores. The statistic calculation unit 105 is responsible for the process of calculating statistics such as the average value, standard deviation, kurtosis, and skewness for each time-series data of the sales quantity of each product in each store. The correlation matrix calculation unit 106 is responsible for the process of calculating the correlation of the sales quantity between stores using the time-series data of the sales quantity of each product in each store. The correlation matrix correction unit 107 is responsible for the process of selecting a correlation matrix from among a plurality of correlation matrices along the time series, selecting a correlation coefficient that is a matrix element from among the selected correlation matrices, and correcting the value of the correlation coefficient. The graph generation unit 108 is responsible for the process of generating a graph that represents stores as nodes and represents the relationships between stores as edges. The attribute data addition unit 109 is responsible for the process of assigning the aforementioned statistics as attribute data to the nodes of the generated graph. The time feature quantity processing unit 110 is used for the processing of the feature quantity of the time-series data of the sales quantity of each product in each store, and in the framework of deep learning, is responsible for the process of enabling the learning and extraction of the feature quantity of the time-series data. The graph feature quantity processing unit 111 is used for the processing of the feature quantity of the generated graph, and in the framework of deep learning, is responsible for the process of enabling the learning and extraction of the feature quantity of the graph. The feature quantity integration processing unit 112 is responsible for the process of performing learning regarding the prediction of the sales quantity of products in the framework of deep learning using both the graph feature quantity and the time feature quantity of the time-series data, and the process of predicting the sales quantity of products based on the learning result. The prediction result output unit 113 is responsible for the process of outputting the result predicted by the deep learning unit 114, such as displaying it on the screen.

[0017] Next, with reference to FIG. 2, an example of the hardware configuration of the sales quantity prediction system will be described.

[0018] First, the sales quantity measurement unit 101 will be described. As an example, the sales quantity measurement unit 101 can be constituted by a computer arranged in each store. In the sales quantity measurement unit 101, each hardware component is connected via a bus, and the sales quantity measurement unit 101 includes a control unit 201, a storage device 202, and a communication unit 203. The control unit 201 can be, for example, a CPU (Central Processing Unit), and is the main body that realizes predetermined processing by executing a program for operating the sales quantity measurement unit 101. Here, data such as programs executed by the control unit 201 of the sales quantity measurement unit 101 is stored in the storage device 202, and the storage device 202 can be appropriately configured using an HDD (Hard Disk Drive), a ROM (Read Only Memory), or the like. Further, the sales quantity measurement unit 101 may include a RAM (Random Access Memory) 204, and the control unit 201 may read data into the RAM 204 and perform processing. The communication unit 203 is an interface used for communication with the controller unit 102, and communication with the controller unit 102 is performed via the communication unit 203.

[0019] Next, the controller unit 102 will be described. In the controller unit 102, each hardware component is connected via a bus, and the controller unit 102 includes a processor 251, a storage device 252, and a communication unit 253. The processor 251 is the main body that realizes predetermined processing by executing a program stored in the storage device 252. The storage device 252 can be appropriately configured using a ROM, an HDD, or the like, and appropriately stores data used for processing such as programs. Further, the controller unit 102 may include a RAM 254, and the processor 251 may read data into the RAM 254 and perform processing.

[0020] Also, in this embodiment, the controller unit 102 includes a display unit 255. The display unit 255 is configured as an appropriate display device and can display, for example, prediction results. Further, the controller unit 102 may include an input unit 256 configured as an input device used when a user performs a predetermined operation (for example, when the user starts a learning process or outputs a prediction result). However, the input unit 256 may be omitted, and the sales quantity prediction system 1 may be configured to automatically execute a predetermined task without the user's operation in the controller unit 102. Also, the display unit 255 and the input unit 256 may be external components connected to the controller unit 102.

[0021] Next, with reference to FIG. 3, the processing flow in the controller unit 102 will be described. The processing in the controller unit 102 is divided into processing in the learning phase and processing in the prediction phase. Here, first, the processing in the learning phase will be described. FIG. 3 is a flowchart showing an example of the processing flow in the learning phase in the controller unit (an example of feature extraction processing by graph convolution and learning processing of action values).

[0022] When the controller unit 102 starts processing, first, through initialization processing, learning parameters and the like are set to initial values (S301).

[0023] Next, in the measurement data collection process realized by the processor executing the measurement data collection unit 103, the log of the measurement data sent from the sales quantity measurement unit 101 to the controller unit 102 via the Internet or the like is collected (S302).

[0024] This log of measurement data records the history of the specified period for each of the product classification of the sold products, the model number unique to each product, the sales quantity, and the date and time of sale. Here, an example of the obtained time-series data is shown in FIG. 4. In FIG. 4, the horizontal axis corresponds to the date and time, and the vertical axis corresponds to the sales quantity. In this example, when focusing on a certain product in a certain product classification, an example of the sales quantity data obtained for each store is shown. In the figure, for simplicity, the sales quantity data corresponding to five stores are represented by 401 to 405 respectively (that is, the data regarding the sales quantity of a certain product common to the five stores are represented by 401 to 405 respectively), but in reality, there are more stores, and it is assumed that similar time-series data are obtained for each product handled in the store.

[0025] Next, in the measurement data interpolation process realized by the processor 251 executing the measurement data interpolation unit 104, when the product to be predicted for the sales quantity is not commonly handled among stores, a process of interpolating the time-series data of the sales quantity is performed (S303). The flow of this process is shown in FIG. 5. FIG. 5 is a flowchart showing an example of the interpolation process of time-series data.

[0026] Here, in this example, it is assumed that only one certain store does not handle the product for which the sales quantity is to be predicted, and the other stores do. In this situation, first, a process of obtaining a group of stores with a similar product composition based on the product classification data included in the measurement data is executed (S501). This group of stores is obtained by searching for stores where the number of matching items in the product classification exceeds a threshold value.

[0027] Next, among a group of stores with similar product configurations, a process is performed to obtain the average value of the time-series data of the sales quantity of the product for which the sales quantity is to be predicted (S502). Next, a process is performed to obtain the same number of noises as the time-series data (S503). Here, as an example, the noise can be a statistic related to the time-series data of the sales quantity of the product for the group of stores obtained in S501 above, and the statistic of the time-series data of the product for each store is obtained as the noise. Here, the noise is, for example, a value probabilistically obtained within the range of a normal distribution obtained by calculating the standard deviation of the time-series data.

[0028] Next, a process is performed to generate new time-series data by adding each of the obtained noises to each value of the time-series data that has already been obtained (that is, the average value of the time-series data obtained in S502). In this way, interpolation is performed by assigning the generated time-series data as the time-series data of stores that do not handle the product to be predicted (S504).

[0029] In the statistic calculation process realized by the processor 251 executing the statistic calculation unit 105, a process is performed to calculate statistics such as the average value, standard deviation, kurtosis, and skewness for each time-series data of the sales quantity of each product in each store (S304).

[0030] Next, in the correlation matrix calculation process realized by the processor 251 executing the correlation matrix calculation unit 106, a process is performed to calculate the correlation of the sales quantity between stores using the time-series data of the sales quantity of each product in each store (S305). This process is to obtain the correlation coefficient for all combinations between stores for the sales quantity for each product (for each product with a unique model number) and generate a correlation matrix. By performing the process of generating the correlation matrix for each fixed period of the time-series data of the sales quantity, a plurality of correlation matrices for each fixed period and for each product are obtained.

[0031] Next, a correlation coefficient correction process realized by the processor 251 executing the correlation matrix correction unit 107 is performed (S306). The flow of this process is shown in FIG. 6. FIG. 6 is a flowchart showing an example of the correlation coefficient correction process (an example of overfitting reduction processing by edge cutting).

[0032] Here, first, a plurality of correlation matrices for each fixed time and for each product obtained in the correlation matrix calculation process are input and passed to the correlation matrix selection process. Here, a plurality of correlation matrices are randomly selected so as to be a specified ratio with respect to the total number of correlation matrices (S601).

[0033] Next, as the correlation coefficient selection process, in each of the selected plurality of correlation matrices, a plurality of correlation coefficients are randomly selected so as to be a specified ratio with respect to the total number of correlation coefficients (S602).

[0034] Next, for each of the selected plurality of correlation coefficients in the selected plurality of correlation matrices, a determination process of the magnitude of the correlation coefficient is performed (S603), and when the absolute value thereof is equal to or greater than the threshold value, a correlation coefficient change process of setting the correlation coefficient to 0 is performed (S604). In the next graph generation process (that is, in S307), since an edge drawing process is performed according to the magnitude of the correlation coefficient, this operation of setting the correlation coefficient to 0 corresponds to an operation of deleting an edge, but this is intended to prevent overemphasizing the relationship between specific stores in the learning to be described in detail later.

[0035] In the graph generation process realized by the processor 251 executing the graph generation unit 108, a process of generating a graph is performed (S307), where stores are used as nodes and the relationships between stores are represented by edges. Based on the fact that the correlation coefficients included in the already calculated correlation matrix represent the correlation of sales quantities between stores, when the absolute value of the correlation coefficient is greater than or equal to the threshold, it is regarded that there is a relationship between stores, and by performing an operation (process) of connecting edges between the corresponding store nodes, the processor 251 generates a graph. Here, an example of a graph representing the relationship between stores when focusing on a certain product is shown in FIG. 7. Here, the nodes representing stores are denoted by reference numeral 701, and the edges indicating that there is a relationship between stores are represented by reference numeral 702. Also, in the graph, the length of the edge is related to the magnitude of the correlation coefficient.

[0036] Next, in the attribute data addition process realized by the processor 251 executing the attribute data addition unit 109, a process of assigning the above-mentioned statistic (the statistic obtained in S304) as attribute data to the nodes of the generated graph is performed (S308). That is, this process is performed by associating each data of the already calculated average value, standard deviation, kurtosis, and skewness as data of the statistic of the time-series data of the sales quantity for each product with the node corresponding to the store where the time-series data was obtained.

[0037] Next, the learning process of time-series data and the graph is performed (S309) by the processor 251 executing the program related to the deep learning unit 114. Here, an example of the configuration of the neural network model for this learning is shown in FIG. 8. This model is mainly composed of a time feature processing layer 801, a graph feature processing layer 802, and a feature integration processing layer 803. Also, the feature integration processing layer 803 is composed of a feature combination layer 804, a batch normalization layer 805, a dropout layer 806, and a fully connected layer 807.

[0038] Here, among the data input to the model, the time-series data is input to the temporal feature processing layer 801. Here, the temporal feature processing layer 801 is assumed to be an LSTM (Long Short Term Memory), and it is assumed that by using this, the features of the time-series change in the sales quantity are learned. Note that the processing related to the learning and extraction of the features of the time-series data is realized by the processor 251 executing the temporal feature processing unit 110.

[0039] Also, among the data input to the model, the graph is input to the graph feature processing layer 802. Here, the graph feature processing layer 802 is assumed to be a GCN (Graph Convolutional Network), and it is assumed that the features of the graph are learned. Note that the processing related to the learning and extraction of the features of the graph is realized by the processor 251 executing the graph feature processing unit 802.

[0040] This processing is performed by a convolution operation that multiplies and adds weights to the attribute data of the nodes connected to each node forming the graph according to the framework of the GCN. For example, in the graph of FIG. 7, focusing on node 703, the attribute data of the nodes connected to it (the nodes inside the range 704) is convolved with the attribute data of node 703. As a result, the attribute data of the nodes connected to node 703 (the nodes other than node 703 inside the range 704) (node attribute 705 in FIG. 7) is reflected in the attribute data of node 703, and the features around node 703 are obtained. When such a convolution process is performed once, the attribute data of the nodes directly connected to the node being focused on is convolved. However, by connecting a plurality of graph feature processing layers 802 and performing the convolution process a plurality of times, the attribute data of nodes that are not directly connected to the node being focused on and are distant is also convolved, and features in a wider range centered on the node being focused on are obtained.

[0041] The features obtained by the time feature quantity processing layer 801 and the graph feature quantity processing layer 802 are input into the feature quantity integration processing layer 803, and first, they are combined in the feature quantity combination layer 804. The combined feature quantities pass through the batch normalization layer 805, the dropout layer 806, and the fully connected layer 807 to output the predicted value of the sales quantity. In the learning phase, the difference (error) between the predicted value of the sales quantity and the actual value of the sales quantity obtained from the measurement data is obtained, and the parameters of the weights of the neural network included in the model are updated so that this error becomes smaller. In the learning process, as an example, common time series data and graphs are used for learning. Also, the processing related to learning and prediction is realized by the processor 251 executing the feature quantity integration processing unit 112.

[0042] Through the above processing, the learning progresses, and in the learning completion determination process realized by the processor executing an appropriate program, when the change rate of the error becomes smaller than the threshold value, it is determined that the learning is completed; otherwise, it is determined that the learning is not completed (S310).

[0043] When the learning is completed, it enters the prediction phase. In this case, the processing is almost the same as the flow of the processing in the learning phase of FIG. 3, but the measurement data up to the day before the day for which the sales quantity is to be predicted is obtained, and the sales quantity is predicted using the learned model. A more detailed difference from the processing flow in FIG. 3 is that in the measurement data collection process (that is, the process of S302), the data of the day before the day to be predicted is received. Also, in the learning process of the time series data and the graph, the predicted value of the sales quantity is output by inputting the time series data and the graph into the model without performing learning (that is, instead of the processing related to S309~S310, the predicted value is output using the learned model). Note that the processing related to the output of the predicted value of the sales quantity is realized by the processor 251 executing the prediction result output unit 113.

[0044] As described above, by learning the time-series data and the feature quantities of the graph, it becomes possible to predict the sales volume. Therefore, according to the sales volume prediction system 1, the relationships between the data of multiple stores can be grasped comprehensively, and a more appropriate prediction of the sales volume of products can be made. And based on the prediction of the sales volume of products with the relationships between the data grasped comprehensively, the user can make an appropriate business plan. Thus, the sales volume prediction system 1 can also contribute to resource savings (for example, reduction of product loss).

[0045] Here, the processor 251 can cause the display unit 255 to display data regarding the predicted value of the sales volume of products and the like by executing an appropriate program (for example, the prediction result output unit 113). The display mode is not particularly limited. For example, a display may be made in a graph format, a table format, or the like so that the change in the sales volume over time can be understood. Also, a display that emphasizes the time zone in which a predetermined sales volume can be expected may be made. Further, the processor 251 may execute output or display to a display device or the like connected to the outside of the system.

[0046] Although the sales volume prediction system 1 of this embodiment has been described in detail, the present invention is not limited to the embodiment and includes various modifications. For example, it is possible to add, delete, or replace a part of the configuration of the embodiment with other configurations.

[0047] For example, in the above description, it is assumed that the sales volume measurement unit 101 is installed in each store and the controller unit 102 is installed in a center that manages the entire plurality of stores, but the controller unit 102 may be in any of the stores. Also, a configuration may be adopted in which the controller unit 102 is installed in each store and each store collects data of other stores and learns and predicts the sales volume.

[0048] For example, the sales quantity measurement unit 101 may be a computer that measures data such as the sales quantity of products and transmits the measured data to the controller unit 102. As an example, it may be a computer having a server function for accumulating or collecting data in the store.

[0049] Also, as an example, the sales quantity measurement unit 101 may be a computer (for example, a business personal computer) that measures data such as the sales quantity of products using the data of that day directly input by a store clerk or the like after closing the store, etc., and transmits the measured data to the controller unit 102.

[0050] The timing at which the sales quantity measurement unit 101 transmits data and the timing at which the controller unit 102 collects data may be set as appropriate. For example, data transmission and reception may be performed at regular intervals, or once a day or every few days.

[0051] Although an example in which the sales quantity measurement unit 101 measures data such as the sales quantity of products and transmits the result to the controller unit 102 has been described, measurement of data such as the sales quantity of products may be executed in the controller unit 102 using the data transmitted from the sales quantity measurement unit 101. In this case, the configurations, programs, etc. of the sales quantity measurement unit 101 and the controller unit 102 may be changed as appropriate.

[0052] In the embodiment, an example of processing a graph in which attribute data is added only to nodes among nodes and edges by convolution has been shown, but this is not the only case. In the processing by the attribute data addition unit 109 of the sales quantity prediction system 1, attribute data may also be added to edges, and in the processing by the graph feature quantity processing unit 111, a graph in which attribute data is also added to edges may be processed by convolution.

[0053] In the embodiment, an example of using convolution processing by GCN for feature extraction has been shown, but this is not the only case. Convolution processing other than GCN may be used.

[0054] In addition, a pooling layer or a dropout layer may be combined with the temporal feature amount processing layer 801 or the graph feature amount processing layer 802.

[0055] The controller unit 102 can be configured to have the same functions as a PC (personal computer). The controller unit 102 includes, for example, a CPU, a memory, a communication device, a user interface for handling basic operations and displaying processing results, basic hardware such as a power supply and wiring, and basic software such as an OS, various firmware, and drivers for controlling these components, and can be assumed to be equipped with what is necessary for operating each unit.

[0056] The control unit 201 and the processor 251 can be a CPU, but any entity that executes predetermined processing may be used, and other semiconductor devices (for example, GPU: Graphics Processing Unit) may also be used.

[0057] Regarding each unit (103 to 113) of the controller 102, although software implementation is assumed here, all or part of them may be implemented as hardware. Also, regarding each unit (103 to 113), if communication is possible, it may be located, for example, in a remote location. Further, the hardware and software constituting the above units (103 to 113) may be selected as appropriate according to the embodiment.

[0058] The learned model may be generated at a remotely located place where communication is possible, and the controller unit 102 may use the learned model generated at a remotely located place where communication is possible to predict the sales volume of a product. Further, as an example, the controller unit 102 may download and use the learned model generated by the same method described in this specification by an appropriate method.

[0059] In addition, in the embodiment, as an example, a system for predicting the sales volume of products in each store forming a chain store has been described. However, as long as it is a system that predicts time-series data based on time-series data obtained at multiple bases, it may be a system that handles other applications.

[0060] A learned model may be generated using time-series data having an appropriate relationship. For example, from the perspective of the store side, a learned model may be generated using time-series data with a common store-opening area. Thereby, based on the perspective of the store side, it is possible to predict more comprehensively the sales volume of products in a predetermined store-opening area.

[0061] For example, by limiting it to the time-series data of stores with a common store-opening area in the Kanto region, it becomes possible to predict more comprehensively the sales volume of products in the Kanto region. Here, an example where the common store-opening area is in the unit of local division has been described, but the range of the common store-opening area is not limited to this example. For example, data common in units such as region, division, and state may be used as the store-opening area, or data common in units such as municipality, prefecture, and city, town, and village may be used.

[0062] Also, instead of or in addition to this condition, from the perspective of the purchasing layer (for example, gender, age, occupation, etc.), a learned model may be generated using time-series data with a common purchasing layer. Thereby, based on the perspective of the purchasing layer, a more comprehensive prediction can be made.

[0063] In the interpolation process of data on products not commonly handled executed in S303, interpolation may be executed based on data having the same scale. By performing interpolation using such data, more appropriate interpolation can be carried out. Here, in S303, for example, conditions may be set such that interpolation is performed based on data of stores with close distances, data of stores in the same opening area of the store, data of stores with similar purchasing layers (as an example, data of stores with similar age and gender distributions, etc.). That is, in S303, the complement processing may be performed based on data common in at least one of the perspectives of the distance between stores, the opening area of the store, and the purchasing layer.

[0064] The sales quantity prediction system 1 can be used, for example, as an example, for chain stores related to convenience stores, drugstores, dealers of automobiles, clothing stores, etc., but is not limited thereto.

Explanation of Signs

[0065] 1 Sales quantity prediction system 101 Sales quantity measurement unit 102 Controller unit 103 Measurement data collection unit 104 Measurement data interpolation unit 105 Statistical quantity calculation unit 106 Correlation matrix calculation unit 107 Correlation coefficient correction unit 108 Graph generation unit 109 Attribute data addition unit 110 Temporal feature processing unit 111 Graph feature processing unit 112 Feature integration processing unit 113 Prediction result output unit

Claims

1. comprising a processor, the processor uses a learned model that predicts the sales quantity of products in a target store based on data related to the sales quantities of products in each store, where the products handled between stores may be partially different, the learned model is generated based on data of products common to each store when the products handled between stores are partially different, the learned model obtains a correlation matrix regarding the sales quantities of the same product between stores and a statistic regarding the sales quantity of the product in the store, generates a graph in which the correlation coefficients of the correlation matrix are associated with edges with the stores in the correlation matrix as nodes, assigns the statistic regarding the store as attribute data to the nodes, and is generated based on the feature amount extracted by convolutional the graph assigned with the statistic, A sales quantity prediction system characterized by the above.

2. comprising a processor, the processor uses a learned model that predicts the sales quantity of products in a target store based on data related to the sales quantities of products in each store, where the products handled between stores may be partially different, the learned model when the products handled between stores are partially different, is generated using data in which data regarding the sales quantities of products not common between stores is interpolated using data regarding the sales quantities of the same products common between other stores, the learned model Obtain a correlation matrix regarding the sales quantities of the same product among stores and statistical quantities regarding the sales quantity of the product in the store, generate a graph in which the correlation coefficients of the correlation matrix are associated with edges with the stores in the correlation matrix as nodes, assign the statistical quantities regarding the store as attribute data to the nodes, and generate based on the feature quantities extracted by folding the graph to which the statistical quantities are assigned. A sales quantity prediction system characterized by the above.

3. The sales quantity prediction system according to claim 1, wherein The learned model is generated by randomly selecting the correlation matrix, randomly selecting a correlation coefficient from the selected correlation matrix, and performing learning while executing a process of setting it to 0 when the selected correlation coefficient is equal to or greater than a threshold value. A sales quantity prediction system characterized by the above.

4. The sales quantity prediction system according to claim 2, wherein The learned model is generated by randomly selecting the correlation matrix, randomly selecting a correlation coefficient from the selected correlation matrix, and performing learning while executing a process of setting it to 0 when the selected correlation coefficient is equal to or greater than a threshold value. A sales quantity prediction system characterized by the above.

5. The sales quantity prediction system according to claim 1, wherein the processor displays information regarding the predicted result on a display unit. A sales quantity prediction system characterized by the above.

6. The sales quantity prediction system according to claim 2, wherein the processor displays information regarding the predicted result on a display unit. A sales quantity prediction system characterized by the above.

7. A sales quantity prediction method performed using a processor, comprising: (1) Generate a trained model for predicting the sales volume of products based on data regarding the sales volume of products in each store, where the products handled between stores may be partially different. (2) Use the trained model to predict the sales volume of products in the target store. The trained model is generated based on data of products common to each store when the products handled between stores are partially different. The trained model obtains a correlation matrix regarding the sales volume of the same product between stores and a statistic regarding the sales volume of the product in the store, generates a graph in which the correlation coefficients of the correlation matrix are associated with edges with the stores in the correlation matrix as nodes, assigns the statistic regarding the store as attribute data to the nodes, and is generated based on the feature amount extracted by folding the graph to which the statistic is assigned. A sales volume prediction method characterized by the above.

8. A sales volume prediction method performed using a processor, (1) Generate a trained model for predicting the sales volume of products based on data regarding the sales volume of products in each store, where the products handled between stores may be partially different. (2) Use the trained model to predict the sales volume of products in the target store. The trained model is generated using data in which data regarding the sales volume of products not common between stores is interpolated using data regarding the sales volume of the same products common between other stores when the products handled between stores are partially different. The trained model Obtain a correlation matrix regarding the sales quantities of the same product among stores and statistical quantities regarding the sales quantity of the product in the store, generate a graph in which the correlation coefficients of the correlation matrix are associated with edges with the stores in the correlation matrix as nodes, assign the statistical quantities regarding the store as attribute data to the nodes, and generate based on the feature quantities extracted by folding the graph to which the statistical quantities are assigned. A sales quantity prediction method characterized by the above.

9. A program for causing the processor to execute the sales quantity prediction method according to Claim 7.

10. A program for causing the processor to execute the sales quantity prediction method according to Claim 8.

Images