Information processing device, information processing method, and information processing program

Abstract

This invention provides an information processing device, an information processing method, and an information processing program that support ordering products, enabling the selection of an ordering logic to be used in a situation where multiple ordering logics with different logics exist. [Solution] The method involves acquiring product identification information for identifying a product and logic identification information for identifying the ordering logic used to order that product as a training dataset, training a machine learning model using the acquired training dataset so that when product identification information is input, it outputs the logic identification information used to order that product, inputting product identification information into the trained machine learning model S110, and when the trained machine learning model outputs the logic identification information, executing the process to order the product using the ordering logic corresponding to the logic identification information S130.

Description

Technical Field

[0004]

[0001] The present invention relates to an information processing apparatus, an information processing method, and an information processing program.

Background Art

[0002] The following order processing apparatus is known. In this order processing apparatus, based on the specified inventory data and the store inventory data, for the target product, when the current inventory quantity for each store is below the order point, the required quantity of the product is calculated as the specified inventory quantity - the current inventory quantity, and the supplier inventory of the target product in the supplier inventory data is allocated the order quantity according to the required quantity of each store in the priority order of each store in the specified inventory data, and order confirmation data including the store, the product, and the order quantity is automatically generated to perform automatic ordering of the product (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

[0006] According to the present invention, a mechanism can be provided to support the selection of an order placement logic to be used in a situation where multiple order placement logics with different logics exist. [Brief explanation of the drawing]

[0007] [Figure 1] This is a block diagram showing the configuration of one embodiment of the information processing device 100. [Figure 2] This is a flowchart illustrating the processing flow during the training phase of a machine learning model. [Figure 3] This flowchart illustrates the processing flow during the inference stage using a pre-trained machine learning model. [Modes for carrying out the invention]

[0008] Figure 1 is a block diagram showing the configuration of one embodiment of the information processing device 100 in this embodiment. The information processing device 100 may be a personal computer or a server, and includes a communication module 101, a control device 102, and a recording device 103.

[0009] The communication module 101 includes a module for connecting the information processing device 100 to a communication line such as a LAN or the Internet, either wirelessly or via a wired connection. In this embodiment, the information processing device 100 can communicate with other devices by connecting to a communication line via the communication module 101.

[0010] The control device 102 consists of a CPU, memory, and other peripheral circuits, and controls the entire information processing device 100. The memory that makes up the control device 102 is a volatile memory such as SDRAM. This memory is used as work memory for the CPU to load programs when executing programs, and as buffer memory for temporarily recording data. For example, data received via the communication module 101 is temporarily recorded in the buffer memory.

[0011] The recording device 103 is a recording device for recording various data stored by the information processing device 100, as well as program data for execution by the control device 102. For example, an HDD (Hard Disk Drive) or SSD (Solid State Drive) may be used. The program data to be recorded in the recording device 103 is provided on a recording medium such as a CD-ROM or DVD-ROM, or via a network. By installing the program data obtained by the user into the recording device 103, the control device 102 becomes able to execute the program.

[0012] In a situation where an ordering logic for automating the ordering of goods is being used, the information processing apparatus 100 in the present embodiment provides a mechanism for selecting an optimal ordering logic for each product when there are multiple ordering logics with different processing contents, procedures, and methods, that is, different logics. In the present embodiment, the ordering of goods is assumed to be in various scenarios such as, for example, ordering goods from a retail store to a logistics company or wholesaler, ordering goods from a logistics company or wholesaler to a manufacturer, and ordering goods from a manufacturer to a raw material manufacturer. Also, the retail store is assumed to include a physical store and a virtual store on the Internet.

[0013] Generally, there are countless ordering logics for automatically ordering goods, each with its own characteristics. For example, there are various ordering logics with characteristics in terms of processing content, procedure, and method, such as an ordering logic that determines an optimal order quantity considering the seasonality of the product, an ordering logic that determines an optimal order quantity considering the intermittency of the product's sales, and an ordering logic that determines an optimal order quantity considering the implementation status of sales promotion activities.

[0014] In a situation where there are countless ordering logics with such characteristics, the results such as the out-of-stock rate, waste rate, and sales of the product may change depending on which ordering logic is adopted. This is because each ordering logic prioritizes different factors when determining the order quantity, resulting in a change in the ordered quantity of the product depending on which ordering logic is adopted. In view of such a situation, the information processing apparatus 100 in the present embodiment aims to provide a mechanism for assisting in the selection of an optimal ordering logic by determining which ordering logic should be used for each product in a situation where multiple ordering logics exist. Hereinafter, the processing executed by the information processing apparatus 100 in the present embodiment will be described.

[0015] In the information processing apparatus 100 according to the present embodiment, when inputting product identification information for identifying a product, a method of selecting an optimal ordering logic for each product is adopted using a machine learning model configured to output, as an inference result, logic identification information for identifying an ordering logic to be used for ordering the product. Note that, as the product identification information for identifying a product, for example, a product ID, a product name, a JAN code, or the like is used. Further, as the logic identification information for identifying an ordering logic, for example, an identification ID for identifying an ordering logic is used. In the present embodiment, an identification ID is assigned to each of the programs of a plurality of ordering logics, and the control device 102 can specify the program of the ordering logic based on the identification ID output from the machine learning model.

[0016] Note that, the program for ordering a product using an ordering logic is recorded in a recording device that can be connected via the recording device 103 or a communication line, and the control device 102 can automatically order a product by acquiring and executing the ordering logic with the identification number output from the machine learning model from the recording destination.

[0017] A learning process for machine learning the machine learning model used in the present embodiment will be described. In the present embodiment, the machine learning model learns, using past data, data indicating which logic achieved the best results when each product was ordered as teacher data. The teacher data is a data set in which input data and a correct label are set. In the present embodiment, for example, data in which product identification information is used as input data and the identification ID of the ordering logic that achieved the best results for that product in the past is used as the correct label is used as teacher data. Then, this teacher data is used as a learning data set to cause the machine learning model to learn. Note that, the method for determining the ordering logic that achieved the best results is not particularly limited, but in the present embodiment, for example, a logic with a low out-of-stock rate, a logic with a low scrap rate, a logic with high sales, or the like is regarded as a good result, and the logic that achieved the best results based on at least one of these results is specified as the best logic.

[0018] The procedure for creating training data is as follows. Note that this example describes creating training data using the information processing device 100, but the training data may be created using other devices. The control device 102 uses pre-prepared historical data to perform simulations using multiple ordering logics for each product. The control device 102 then calculates KPIs such as sales, stockout rate, and waste rate, and creates the correct label by identifying the best logic that achieved the best results for that product based on the calculation results. The control device 102 uses the product identification information of the simulated product as input data, and creates training data using the identification ID of the best logic obtained from the simulation results as the correct label. The control device 102 records the created training data in the recording device 103. This makes it possible to train a machine learning model using the training data recorded in the recording device 103.

[0019] The control device 102 acquires the training data created in this way from the recording device 103 as a training dataset, inputs it into the model to be trained, trains the machine learning model, and records data for using the trained machine learning model. This allows the control device 102 to perform processing using the recorded trained machine learning model. The destination for recording the trained machine learning model is not particularly limited. For example, the control device 102 may record the trained machine learning model in the recording device 103, or it may record it in another device that can communicate via a communication line. In the following description, it is assumed that the trained machine learning model is recorded in the recording device 103.

[0020] In this embodiment, when product identification information of the product to be ordered is input to the trained machine learning model, the trained machine learning model outputs logic identification information of the optimal ordering logic for ordering that product as an inference result, so that the optimal ordering logic can be selected for each product. In other words, the control device 102 can input product identification information of the product to be ordered into the trained machine learning model and identify the logic identification information of the optimal ordering logic based on the inference result output from the trained machine learning model. The control device 102 can order the product using the optimal ordering logic by executing the ordering logic program corresponding to the logic identification information output from the trained machine learning model, that is, the ordering logic program to which the identification ID output from the trained machine learning model has been assigned.

[0021] Figure 2 is a flowchart showing the processing flow during the training phase of a machine learning model executed by the information processing device 100 in this embodiment. The processing shown in Figure 2 is executed by the control device 102 as a program that is started at the timing when the machine learning model is trained. Note that the timing of training the machine learning model is not particularly limited.

[0022] In step S10, the control device 102 acquires the training data recorded in the recording device 103 as a training dataset, as described above. Then, the process proceeds to step S20.

[0023] In step S20, the control device 102 inputs the acquired training dataset into the machine learning model to train the machine learning model. After that, the process proceeds to step S30.

[0024] In step S30, the control device 102 determines whether the training of the machine learning model has finished. The control device 102 may determine that the training of the machine learning model has finished, for example, when the operator instructs the machine learning model to finish training, or when training of all training data has been completed. If the determination in step S30 is negative, the process returns to step S10. On the other hand, if the determination in step S30 is positive, the process proceeds to step S40.

[0025] In step S40, the control device 102 records the data for using the trained machine learning model to a recording destination, such as the recording device 103, as described above. After that, the process ends.

[0026] Figure 3 is a flowchart showing the processing flow during the inference stage using a trained machine learning model executed by the information processing device 100 in this embodiment. The processing shown in Figure 3 is executed by the control device 102 as a program that is started at the start timing of the inference process. Note that the start timing of the inference process is not particularly limited.

[0027] In step S110, the control device 102 inputs the product identification information of the product to be ordered into the trained machine learning model. Then, the process proceeds to step S120.

[0028] In step S120, the control device 102 determines whether the trained machine learning model has output the inference result data described above, i.e., the identification ID of the optimal ordering logic to be used to order the product. If the determination in step S120 is positive, the process proceeds to step S130.

[0029] In step S130, the control device 102 executes the process of ordering goods by running the program of the ordering logic to which the inference result output from the trained machine learning model, i.e., the identification ID of the optimal ordering logic to be used to order the goods, has been assigned, as described above. After that, the process is terminated.

[0030] According to the embodiments described above, the following effects and advantages can be obtained. (1) The control device 102 acquires product identification information for identifying products and logic identification information for identifying the ordering logic used to order those products as a training dataset. Using the acquired training dataset, the control device trains a machine learning model so that when product identification information is input, it outputs the logic identification information used to order that product. This allows the machine learning model to be trained to output the identification information of the optimal ordering logic used to order products.

[0031] (2) The machine learning model was defined as a model that was trained using data that showed which logic resulted in the best outcome when ordering each product, based on past data. This allows the machine learning model to learn the training data so that it can infer the optimal ordering logic for each product based on past data.

[0032] (3) The training data was a dataset in which product identification information of products was used as input data, and the logic identification information of the best logic obtained from the results of simulations using historical data was used as the ground truth label. This makes it possible to prepare training data that allows the machine learning model to infer the optimal ordering logic for each product based on the simulation results using historical data.

[0033] (4) The machine learning model selects the appropriate ordering logic from among several ordering logics with different logics to order a product, and outputs an identification ID for the selected ordering logic. This helps in selecting which ordering logic to use when there are multiple ordering logics with distinctive logics to order a product.

[0034] (5) The control device 102 receives product identification information to identify a product, and inputs the product identification information to a trained machine learning model that has been trained to output logic identification information to identify the ordering logic used to order that product. When logic identification information is output from the trained machine learning model, the control device 102 executes the process of ordering the product using the ordering logic corresponding to the logic identification information. This makes it possible to perform ordering processing using the optimal ordering logic for each product.

[0035] (6) The control device 102 executes the process of ordering products by running an ordering logic program to which an identification ID output from the trained machine learning model has been assigned. This makes it possible to execute the ordering logic program that is optimal for each product and to perform the ordering process.

[0036] —Revised Version— Furthermore, the information processing device of the above-described embodiment can also be modified as follows. (1) In the embodiment described above, product identification information was used as input data, and the identification ID of the ordering logic that had achieved the best results for that product in the past was used as the ground truth label. An example was described in which a machine learning model was trained using a dataset of this input data and ground truth label as the training dataset. However, the trained machine learning model created in this way may also be trained using input data that shows the characteristics of the product and the ground truth label described above. Here, the input data used to represent the characteristics and circumstances of a product (such as the stability and intermittence of demand) will be data that shows the characteristics of the product. For example, data such as store ID, date, percentage of days with zero sales in the most recent period (zero rate), sales variability, seasonality strength, promotion flag, point multiplier, average lead time, lead time variability (variance), indicators of intermittence and seasonality such as average demand interval, information such as temperature, precipitation, pollen and infectious diseases, holidays and events, store inventory, outstanding orders, DC inventory, remaining time until closing time, simple time series summaries such as lag / moving average / moving variance (7 / 14 / 28), trend slope, shelf capacity / expected order quantity ratio, and case quantity / standard order quantity ratio can be input into the model for training. The trained machine learning model, thus trained, can predict which logic is optimal with probability (confidence), making it possible to automatically determine which logic to choose based on a specific product and sales situation (intermittence, seasonality, presence or absence of promotions, etc.). Furthermore, by providing the context of the specified product at that time when training the learning model, this information can be taken into account. This allows for reinforcement learning that accurately understands the execution environment of the logic and uses it as an indicator to find the optimal solution, and enables learning to correct for discrepancies between the output and reality.

[0037] Furthermore, the present invention is not limited in any way to the configurations in the embodiments described above, as long as the characteristic functions of the present invention are not impaired. [Explanation of Symbols]

[0038] 100 Information Processing Devices 101 Communication Module 102 Control device 103 Recording device

Claims

1. A means for acquiring training data, in which product identification information for identifying a product is used as input data, and training data is obtained as a training dataset, in which logic identification information for identifying the ordering logic that performed best for that product, which is identified by simulating with multiple ordering logics using past data for the product identified by the product identification information, is used as the ground truth label. An information processing apparatus comprising: a learning means that, using the training dataset acquired by the acquisition means, trains a machine learning model to output logic identification information used to order a product when the product identification information is input.

2. In the information processing apparatus according to claim 1, The machine learning model is characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

3. An input means for inputting product identification information into a trained machine learning model, which takes product identification information as input data, simulates using multiple ordering logics on products identified by the product identification information using past data, identifies the logic that performed best for that product, and uses this training data as ground truth labels. The trained model is trained to output the logic identification information of the ordering logic used to order that product when the product identification information is input. An information processing apparatus comprising: an order processing execution means that, when the logic identification information is output from the trained machine learning model, executes a process for ordering goods using the order logic corresponding to the logic identification information.

4. In the information processing apparatus described in claim 3, The machine learning model is characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

5. In the information processing apparatus according to claim 4, The order processing execution means is characterized by executing a program of order logic to which the logic identification information has been assigned, thereby executing a process for ordering goods.

6. The acquisition method involves the steps of obtaining training data as a training dataset, using product identification information to identify a product as input data, performing simulations using multiple ordering logics with past data for the product identified by the product identification information, and then obtaining training data as a ground truth label, which is logic identification information to identify the ordering logic that performed best for that product. An information processing method performed on a computer, comprising the steps of: a learning means using the learning dataset acquired by the acquisition means to train a machine learning model so that when product identification information is input, it outputs logic identification information to be used to order the product.

7. In the information processing method described in claim 6, The machine learning model is characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

8. The input means takes product identification information for identifying a product as input data, and uses training data consisting of logic identification information, which identifies the logic that performed best for that product, identified by performing simulations using multiple ordering logics with past data for the product identified by the product identification information, as the ground truth label. The trained machine learning model is trained to output the logic identification information of the ordering logic used to order that product when the product identification information is input. An information processing method performed on a computer, comprising the step of executing a process for ordering goods using the order logic corresponding to the logic identification information when the logic identification information is output from the trained machine learning model.

9. In the information processing method described in claim 8, The machine learning model is characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

10. In the information processing method described in claim 9, The order processing execution means is characterized by executing a program of order logic to which the logic identification information has been assigned, thereby executing a process for ordering goods.

11. The acquisition procedure involves using product identification information to identify a product as input data, performing simulations using multiple ordering logics with historical data for the product identified by the product identification information, and obtaining training data as a training dataset, with logic identification information used as the ground truth label to identify the ordering logic that performed best for that product. An information processing program for causing a computer to execute a learning procedure that uses the training dataset acquired in the acquisition procedure to train a machine learning model so that when the product identification information is input, it outputs the logic identification information used to order that product.

12. In the information processing program described in claim 11, The machine learning model is an information processing program characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

13. An input procedure for inputting product identification information into a trained machine learning model, which is trained to output the logic identification information of the ordering logic used to order a product, using product identification information as input data, and logic identification information, which is the ground truth label for identifying the logic that performed best for that product, identified by simulating with multiple ordering logics using past data for the product identified by the product identification information, as training data, and which is trained to output the logic identification information of the ordering logic used to order that product when the product identification information is input, and the trained machine learning model is trained to output the logic identification information of the ordering logic used to order that product, and the product identification information is input as input data, and the product identification information is used as training data, which is trained to identify the logic that performed best for that product, identified by simulating with multiple ordering logics using past data for the product identified by the product identification information, and the logic identification information of the ordering logic used to order the product is used as training data, and the product identification information is used as input data identified by the product identification information, and the logic identification information of the ordering logic used to order the product is used as training data, and the product identification information is used An information processing program that causes a computer to execute an order processing execution procedure, which, when the logic identification information is output from the trained machine learning model, uses the order logic corresponding to the logic identification information to execute the process of ordering goods.

14. In the information processing program described in claim 13, The machine learning model is an information processing program characterized by selecting an ordering logic to be used to order goods from among multiple ordering logics with different logics, and outputting the logic identification information of the selected ordering logic.

15. In the information processing program described in claim 14, The order processing execution procedure is an information processing program characterized by executing a process for ordering goods by executing a program of order logic to which the logic identification information has been assigned.

Images