Catalog creation support device, catalog creation support method and program

Abstract

The goal is to provide technology to support catalog creation using machine learning models. [Solution] One aspect of the present disclosure relates to a catalog creation support device having a model processing unit that acquires sales forecast values ​​from product candidate data for product candidates at each listing position using a demand forecast model trained on purchase history data including product listing positions in a catalog and actual sales values ​​for the products, and a product placement unit that determines product candidates to be listed at each listing position in the catalog based on the sales forecast values.

Description

[Technical Field]

[0001] The present disclosure relates to a catalog creation support device, a catalog creation support method, and a program. [Background technology]

[0002] A catalog contains many products on various pages and in various positions, and the position of each product in the catalog is determined mainly by the creator or editor. Technologies have been proposed to reduce the labor required for this manual catalog creation work. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-022108 Summary of the Invention [Problem to be solved by the invention]

[0004] With the recent advancement of AI (Artificial Intelligence) technology, machine learning models have been used in various technical fields. There is a demand for technology that uses machine learning models to support the creation of magazines, catalogs, books, etc.

[0005] The objective of the present disclosure is to provide a technology for supporting catalog creation using machine learning models. [Means for solving the problem]

[0006] One aspect of the present disclosure relates to a catalog creation support device having a model processing unit that acquires sales forecast values ​​from product candidate data of product candidates at each listing position using a demand forecast model trained on purchase history data including product listing positions in a catalog and actual sales values ​​of the products, and a product placement unit that determines product candidates to be listed at each listing position in the catalog based on the sales forecast values. [Effects of the Invention]

[0007] According to the present disclosure, a technique for supporting catalog creation using machine learning models can be provided. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a schematic diagram illustrating a catalog creation assistance process according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a block diagram illustrating a hardware configuration of a catalog creation support device according to an embodiment of the present disclosure. [Figure 3] FIG. 3 is a block diagram illustrating a functional configuration of a catalog creation support device according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a diagram showing the placement of items on a catalog page according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram illustrating a value change flag according to one embodiment of the present disclosure. [Figure 6] FIG. 6 is a schematic diagram illustrating a training process for a demand forecasting model according to one embodiment of the present disclosure. [Figure 7] FIG. 7 is a diagram illustrating a data structure of product candidate data according to an embodiment of the present disclosure. [Figure 8] FIG. 8 is a diagram illustrating a data structure of demand forecast data according to an embodiment of the present disclosure. [Figure 9] FIG. 9 is a schematic diagram illustrating an inference process of a demand forecasting model according to an embodiment of the present disclosure. [Figure 10] FIG. 10 is a diagram illustrating demand forecast data according to an embodiment of the present disclosure. [Figure 11] FIG. 11 is a flowchart illustrating a catalog creation support process according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0010] In the following embodiment, a catalog creation support device that supports catalog creation using a machine learning model is disclosed.

[0011] [Summary of this disclosure] First, a catalog creation support process according to an embodiment of the present disclosure will be described. Figure 1 is a schematic diagram illustrating a catalog creation support process according to an embodiment of the present disclosure.

[0012] The catalog creation support device 100 according to the following embodiment estimates the sales forecast value of a product candidate to be listed in a catalog from product candidate data indicating the product information and listing information of the product candidate, using a demand forecast model 50 that has been trained to output sales forecast values ​​indicating predicted values ​​such as sales quantity and sales amount when the product candidate is placed in each listing position in the catalog.

[0013] After estimating the sales forecast value of each product candidate when placed in various placement positions, the catalog creation support device 100 determines the product candidates to be placed in each placement position of the catalog so as to maximize the total sales forecast value of the catalog or catalog page. For example, after acquiring demand forecast data storing catalog placement positions, product candidates, and sales forecast values ​​as shown in FIG. 1, the catalog creation support device 100 identifies the product candidate with the highest sales forecast value for frame position #1 from the demand forecast data and determines to place the identified product candidate in frame position #1. Next, the catalog creation support device 100 identifies the product candidate with the highest sales forecast value for frame position #2 (excluding the product candidate placed in frame position #1) and determines to place the identified product candidate in frame position #2. Similarly, the catalog creation support device 100 identifies product candidates for frame positions #3 to #7 and determines to place the identified product candidate in frame positions #3 to #7. In this manner, the product candidates to be placed in each placement position on the catalog page are determined.

[0014] This makes it possible to determine the product placement that maximizes catalog sales based on the sales forecast value of each product candidate predicted by the demand forecasting model 50 when it is placed in each listing position, thereby assisting catalog creators and editors in creating the catalog.

[0015] Here, the catalog creation support device 100 may have, for example, a hardware configuration as shown in Fig. 2. That is, the catalog creation support device 100 has a storage device 101, a memory device 102, a processor 103, and a communication device 104, which are interconnected via a bus B.

[0016] The programs and / or data that realize the various functions and processes of the catalog creation support device 100 are downloaded to the storage device 101 or memory device 102 via the communication device 104 from an external device and / or a network.

[0017] The storage device 101 is realized by a nonvolatile memory or the like, and stores installed or downloaded programs or data (for example, files, etc.).

[0018] The memory device 102 is realized by a random access memory, a static memory, or the like, and when a program or instruction is activated, reads and stores the program, instruction, data, or the like from the storage device 101. The storage device 101, the memory device 102, and the removable storage medium may be collectively referred to as a non-transitory storage medium.

[0019] The processor 103 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores, and performs various functions and processes of the catalog creation support device 100 in accordance with programs, instructions, data such as parameters required to execute the programs or instructions, etc. stored in the memory device 102.

[0020] The communication device 104 is realized by various communication circuits that execute wired and / or wireless communication processing with external devices, the Internet, a LAN (Local Area Network), a cellular network, or other communication networks.

[0021] However, the above-described hardware configuration is merely an example, and the catalog creation support device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.

[0022] [Catalogue creation support device] Next, a catalog creation support device 100 according to an embodiment of the present disclosure will be described. Fig. 3 is a block diagram showing the functional configuration of the catalog creation support device 100 according to an embodiment of the present disclosure. As shown in Fig. 3, the catalog creation support device 100 includes a model processing unit 110 and a product placement unit 120. Each functional unit of the model processing unit 110 and the product placement unit 120 can be realized by the processor 103 of the catalog creation support device 100 executing a program stored in the memory device 102.

[0023] The model processing unit 110 acquires a sales forecast value from product candidate data of product candidates at each listing position using a demand forecasting model 50 trained with purchase history data including the listing position of the product in the catalog and actual sales values ​​of the product. Specifically, the model processing unit 110 inputs product candidate data indicating the listing position in the catalog and the product candidate into the trained demand forecasting model 50, and acquires from the trained demand forecasting model 50 a forecast value of the sales quantity or sales amount when the product candidate is placed in the listing position in the catalog.

[0024] For example, the product candidate data may be composed of publication information and product information, and specifically may include frame position, frame size, page number, publication timing, price promotion flag, price, and product classification. Here, the frame position, frame size, page number, and publication timing may indicate publication information such as the publication position of the product candidate in the catalog corresponding to the publication timing.

[0025] For example, the catalog page of a catalog published in a certain month may have a frame arrangement as shown in Fig. 4, that is, may be composed of five frame positions a1 to a5. Here, frame positions a1 and a2 have frame size s1, frame position a3 has frame size s2, and frame positions a4 and a5 have frame size s3. Note that the layout pattern regarding how the frame positions are arranged may differ for each page of the catalog, and the product candidate data may further include data items for the layout pattern.

[0026] Furthermore, the price promotion flag may indicate a price description for the product candidate, and may be an identifier indicating a price description for the product candidate, such as "regular price," "price reduction," or "bargain." For example, each price promotion flag fi may have the content shown in FIG. 5. Specifically, the price promotion flag f0 indicates "regular price," the price promotion flag f1 indicates "price reduction," and the price promotion flag f2 indicates "bargain." Furthermore, the price indicates the price of the product candidate, and the product classification indicates the product classification of the product candidate (for example, one or a combination of major product classifications, medium classifications, and minor classifications).

[0027] In this case, the demand forecasting model 50 is trained to output sales forecast values ​​for catalog positions and product candidates identified by frame position, frame size, page number, publication timing, price promotion flag, price, and product classification. However, the product candidate data is not necessarily limited to these data items, and any other appropriate data items indicating catalog positions and product candidates may be used.

[0028] Such a demand forecasting model 50 may be trained, for example, by a training process such as that shown in FIG. 6. Specifically, a training dataset consisting of training data TD#i is stored in a training data database (DB) 20, and a demand forecasting model 40 to be trained is trained using the training dataset by a training device 30 (not shown). Here, the demand forecasting model 40 to be trained may be, for example, a random forest, which performs predictions by combining multiple decision trees. However, the demand forecasting model according to the present disclosure is not limited to this and may be realized by any other appropriate type of machine learning model, such as a deep learning model.

[0029] Each training data TD#i may be composed of, for example, purchase history data #Ri and actual sales value #Si. When the trained demand forecasting model 50 outputs a sales forecast value from frame position, frame size, page number, publication timing, price promotion flag, price, and product category, the purchase history data #Ri may be composed of frame position, frame size, page number, publication timing, price promotion flag, price, and product category in a catalog published in the past, and the actual sales value S#i may be composed of sales results for the purchase history data of the catalog published in the past.

[0030] The training device 30 generates a random forest by generating multiple decision trees from a training dataset consisting of purchase history data #Ri and actual sales values ​​#Si according to any known random forest training algorithm. Specifically, the training device 30 divides the training dataset acquired from the training data DB 20 into multiple subsets, and generates each decision tree of the random forest using each subset according to any known training algorithm. The generated random forest can be used by the model processing unit 110 as a demand forecasting model 50.

[0031] Once each decision tree has been trained in this manner, the model processing unit 110 inputs the product candidate data into the trained random forest and obtains a sales forecast value for the input product candidate data by averaging or taking a majority vote on the output results from each decision tree. The model processing unit 110 provides the obtained sales forecast value to the product placement unit 120 in association with each product candidate data input into the random forest.

[0032] The product placement unit 120 determines product candidates to be placed in each placement position in the catalog based on the sales forecast value. Specifically, the product placement unit 120 acquires sales forecast values ​​such as sales quantity and sales amount when each product candidate is placed in the candidate placement position in the catalog to be created from the model processing unit 110.

[0033] For example, the product placement unit 120 may place the product candidates in a catalog placement position so as to maximize the total sales forecast value of the product candidates in the catalog or on the page of the catalog. Specifically, in order to obtain the sales forecast value for cases in which each product candidate is placed in various placement positions, the product placement unit 120 generates product candidate data corresponding to these cases and provides the generated product candidate data to the model processing unit 110.

[0034] The product candidate data may have a data structure such as that shown in FIG. 7. As shown in FIG. 7, each product candidate data item is composed of the following data items: frame position, frame size, page number, publication timing, price promotion flag, price, and product classification. For example, product candidate data #Cx1 is composed of frame position “a2,” frame size “s1,” page number “p10,” publication timing “2017 / 07,” price promotion flag “f0,” price “1,200 yen,” and product classification “65.” Product candidate data #Cx2 is composed of frame position “a3,” frame size “s2,” page number “p6,” publication timing “2019 / 06,” price promotion flag “f2,” price “4,500 yen,” and product classification “34.” Similarly, each product candidate data #Cxi is composed of these data items.

[0035] When the sales forecast value for each product candidate data is acquired from the model processing unit 110, the product placement unit 120 stores the acquired sales forecast value and the corresponding product candidate data in the demand forecast data. For example, the demand forecast data may have a data structure as shown in FIG. 8. As shown in FIG. 8, each demand forecast data is composed of data items such as frame position, frame size, page number, publication timing, price promotion flag, price, product category, and sales forecast value. Then, in a prediction process using a demand forecast model 50 as shown in FIG. 9, the sales forecast value of each demand forecast data is predicted by the demand forecast model 50 from the frame position, frame size, page number, publication timing, price promotion flag, price, and product category.

[0036] When the publication timing of the catalog is specified by the user, the product placement unit 120 refers to the acquired demand forecast data, identifies the maximum sales forecast value for each listing position on each page of the catalog for the publication timing specified by the user, and determines the product candidate corresponding to the maximum sales forecast value as the product to be placed in that listing position. To achieve this, for example, the product placement unit 120 may determine the product candidate based on a Greedy algorithm, and specifically, may determine the product candidate to be placed in each listing position in the catalog in descending order of the sales forecast values ​​of the product candidates to be placed in each listing position.

[0037] For example, when determining a product to be featured on page "p5" of the catalog for the publication timing "March 2024 issue (2024 / 03)," as shown in Fig. 10, the product placement unit 120 first focuses on the product candidate data related to frame position "a1" on page number "p5" of the demand forecast data, and determines to place the product corresponding to the product candidate with the largest sales forecast value among these product candidate data in frame position "a1" on page number "p5." Specifically, the product placement unit 120 focuses on product candidate data #C70, #C42, #C65, #C12, and #C32 related to frame position "a1" on page number "p5" of the demand forecast data, and determines to place the product corresponding to product candidate #C65 with the largest sales forecast value among these product candidate data in frame position "a1" on page number "p5."

[0038] Next, the product placement unit 120 focuses on the product candidate data related to frame position "a2" on page number "p5" of the demand forecast data, and determines to place the product corresponding to the product candidate with the highest sales forecast value among these product candidate data in frame position "a2" on page number "p5." Here, to avoid duplication of products in the catalog, the determined products exclude products already determined to be included in the catalog. The product placement unit 120 focuses on product candidate data #C46, #C65, and #C36 related to frame position "a2" on page number "p5" of the demand forecast data, excluding products already selected for the catalog, and determines to place the product corresponding to product candidate #C46 with the highest sales forecast value among these product candidate data in frame position "a2" on page number "p5." Similarly, the product placement unit 120 determines products to be included in each placement position on each page of the catalog.

[0039] Furthermore, when the product candidate data includes a layout pattern, the product placement unit 120 calculates the total sales forecast value of the entire catalog page based on the product placement determined for each layout pattern prepared in advance, and determines the layout pattern that maximizes the sales forecast value as the layout pattern for the catalog page.The product placement unit 120 may then apply the product placement determined for the determined layout pattern to the catalog page.

[0040] In the above-described embodiment, product placement in the catalog is determined based on sales forecast values, but the present disclosure is not limited to this, and product placement may be determined by focusing on any indicator that the catalog creator or editor takes into consideration (e.g., profit margin, sales growth rate, etc.).

[0041] According to the above-mentioned catalog creation support device 100, it is possible to determine product placement that maximizes catalog sales based on sales forecast values ​​predicted by the demand forecast model 50, thereby supporting catalog creators and catalog editors in creating catalogs.

[0042] [Catalog creation support process] Next, a catalog creation support process according to an embodiment of the present disclosure will be described. The catalog creation support process can be realized by the catalog creation support device 100, more specifically, by the processor 103 of the catalog creation support device 100 executing a program stored in the memory device 102. Fig. 11 is a flowchart showing the catalog creation support process according to an embodiment of the present disclosure.

[0043] 11, in step S101, the catalog creation support device 100 acquires sales forecast values ​​from product candidate data of product candidates in each listing position using the trained demand forecast model 50. Here, the demand forecast model 50 is trained using a training dataset made up of purchase history data and actual sales values ​​for products listed in catalogs published in the past, and may be realized as, for example, a random forest, without being limited thereto.

[0044] Here, the purchase history data may be composed of publication information and product information, and may be composed of, for example, data items such as frame position, frame size, page number, publication timing, price promotion flag, price, and product classification, and may be created based on the publication information and product information of each product published in a previously published catalog. Similarly, the product candidate data may also be composed of publication information and product information, and may be composed of data items such as frame position, frame size, page number, publication timing, price promotion flag, price, and product classification corresponding to the purchase history data, and the trained demand forecasting model 50, upon receiving the product candidate data as input, outputs a sales forecast value for the input product candidate data.

[0045] The catalog creation support device 100 creates product candidate data for each product to be included in a catalog for a publication timing, such as a publication month, specified by a user, inputs each product candidate data into the trained demand forecasting model 50, and acquires a sales forecast value for the product candidate data. Specifically, the catalog creation support device 100 may create product candidate data corresponding to cases where each product candidate is listed in various positions in the catalog. Upon acquiring the sales forecast value for each product candidate data, the catalog creation support device 100 stores the acquired sales forecast value and the corresponding product candidate data in the demand forecast data.

[0046] In step S102, the catalog creation support device 100 determines product candidates to be placed in each placement position in the catalog based on the sales forecast value. Specifically, the catalog creation support device 100 may place the product candidates in the placement positions in the catalog so as to maximize the total sales forecast value of the product candidates in the catalog or on the page of the catalog. For example, the catalog creation support device 100 identifies the product candidate with the highest sales forecast value in each placement position of the demand forecast data, and determines the product candidate with the highest sales forecast value as the product to be placed in that placement position.

[0047] Specifically, the catalog creation support device 100 may apply a Greedy algorithm to determine product candidates to be placed in each listing position in the catalog in descending order of the sales forecast value of the product candidates to be placed in each listing position. For example, as shown in Fig. 10, the catalog creation support device 100 may extract product candidate data corresponding to the publication timing specified in the demand forecast data, sort the data by page number and frame position, and determine the product candidate corresponding to each page number and frame position that has the largest sales forecast value (excluding products that have been determined to be placed in other listing positions in the catalog) as the product to be placed at that page number and frame position.

[0048] According to the above-described catalog creation support process, it is possible to determine the product placement that maximizes catalog sales based on the sales forecast values ​​predicted by the demand forecasting model 50, thereby supporting catalog creators and catalog editors in creating catalogs.

[0049] Although the examples of the present disclosure have been described in detail above, the present disclosure is not limited to the specific embodiments described above, and various modifications and variations are possible within the scope of the gist of the present disclosure as set forth in the claims. [Explanation of symbols]

[0050] 40,50 Demand forecasting model 100 Catalog creation support device 110 Model processing section 120 Product placement department

Claims

1. a model processing unit that acquires sales forecast values ​​from product candidate data for product candidates at each listing position using a demand forecasting model trained on purchase history data including product listing positions in a catalog and actual sales values ​​of the products; a product placement unit that determines product candidates to be placed in each placement position of the catalog based on the sales forecast value; A catalog creation support device having the above.

2. The catalog creation support device according to claim 1 , wherein the purchase history data and the product candidate data include publication information and product information.

3. The catalog creation support device according to claim 1 , wherein the product placement unit places the product candidates in positions in the catalog so as to maximize a total of predicted sales values ​​of the product candidates in the catalog or on the pages of the catalog.

4. The catalog creation support device according to claim 3 , wherein the product placement unit determines the product candidates to be placed at each placement position in the catalog in descending order of sales forecast values ​​of the product candidates to be placed at each placement position.

5. Obtaining a sales forecast value from product candidate data of product candidates at each listing position using a demand forecasting model trained by purchase history data including product listing positions in a catalog and actual sales values ​​of the products; determining product candidates to be listed in each listing position of the catalog based on the sales forecast value; A catalog creation support method in which the above is executed by a computer.

6. Obtaining a sales forecast value from product candidate data of product candidates at each listing position using a demand forecasting model trained by purchase history data including product listing positions in a catalog and actual sales values ​​of the products; determining product candidates to be listed in each listing position of the catalog based on the sales forecast value; A program that causes a computer to execute the following.

Images