Method, apparatus, and recording medium for analyzing cause of error

Abstract

The present disclosure relates to a method, an apparatus, and a recording medium for obtaining error cause information. The method according to an embodiment of the present disclosure may comprise the steps of: determining an analysis target prediction model, for which a cause of an error between an expected sales amount and an actual sales amount is to be analyzed, among a plurality of prediction models for calculation of expected sales amounts for a plurality of items; identifying an error cause analysis model corresponding to the analysis target prediction model, wherein the error cause analysis model includes one or more determination rules identified on the basis of the corresponding analysis target prediction model; and on the basis of the error cause analysis model corresponding to the analysis target prediction model, obtaining error cause information corresponding to the analysis target prediction model.

Description

Method, device and recording medium for analyzing the cause of errors

[0001] The present disclosure relates to a technique for analyzing the cause of errors in predicting sales opportunities of items in an e-commerce service.

[0002] Recent advancements in communications technology have led to the widespread adoption of e-commerce services, which allow for the trading of goods (items) online. Furthermore, information and communication technologies are being introduced to the transportation, logistics, and inventory management industries related to e-commerce services to improve operational efficiency.

[0003] Meanwhile, accurately estimating sales information is essential for effectively operating e-commerce services. This may require estimating sales opportunities. For example, accurately predicting sales opportunities can reduce lost sales opportunities. Lost sales opportunities can occur, for example, when an item is out of stock due to insufficient inventory. Even if a user is willing to purchase an item, sales cannot be generated if the item is out of stock, resulting in the loss of potential sales opportunities. Therefore, accurately estimating expected sales volumes can avoid stockouts and minimize lost sales opportunities. Conversely, overestimating expected sales volumes increases the likelihood of inventory overstocking, which can lead to inventory storage costs and sales losses.

[0004] However, errors in sales opportunity forecasts can arise from a variety of causes, and failure to properly identify these sources of error can hinder improvements in sales opportunity forecasting methods. Therefore, a solution to this problem is needed.

[0005] A technical problem to be solved through one embodiment of the present disclosure is to more accurately determine the cause of errors in predicting sales opportunities for items in an e-commerce service.

[0006] Another technical problem to be solved through one embodiment of the present disclosure is to more accurately determine a target prediction model requiring analysis of the cause of error among various prediction models used to predict sales opportunities in e-commerce services.

[0007] Another technical challenge to be addressed by one embodiment of the present disclosure is to more accurately identify the risk of stock shortages in e-commerce services.

[0008] The technical problems of the present disclosure are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art of the present disclosure from the description below.

[0009] A method performed by an electronic device according to one embodiment of the present disclosure may include the steps of: determining an analysis target prediction model for analyzing a cause of error between the predicted sales amount and the actual sales amount among a plurality of prediction models that calculate predicted sales amounts for a plurality of items; identifying an error cause analysis model corresponding to the analysis target prediction model, wherein the error cause analysis model includes one or more decision rules identified based on the corresponding analysis target prediction model; and calculating error cause information corresponding to the analysis target prediction model based on the error cause analysis model corresponding to the analysis target prediction model.

[0010] In one embodiment, the step of determining the target prediction model for analysis may include the step of calculating total error information regarding the target prediction model for analysis among the plurality of prediction models based on the total predicted sales corresponding to the plurality of items and the total actual sales corresponding to the plurality of items calculated based on the plurality of prediction models; and the step of determining the target prediction model for analysis among the plurality of prediction models based on the total error information.

[0011] In one embodiment, the total error information may be identified based on a weighted mean absolute percentage error (WMAPE) value between the total expected sales and the total actual sales corresponding to the plurality of items.

[0012] In one embodiment, the step of determining the prediction model to be analyzed may include the steps of calculating error contribution information corresponding to each of the plurality of items; identifying at least some items among the plurality of items for which an error cause is to be analyzed based on the error contribution information; and determining a prediction model corresponding to at least some of the identified items as the prediction model to be analyzed.

[0013] In one embodiment, the step of calculating error contribution information corresponding to each of the plurality of items may include the step of calculating first total error information corresponding to all of the plurality of items; the step of calculating second total error information corresponding to one or more items excluding a target item that is the target of error contribution calculation among the plurality of items; and the step of calculating error contribution information corresponding to the target item based on the first total error information and the second total error information.

[0014] In one embodiment, the prediction model corresponding to at least some of the identified items may be determined based on at least some of the item information, timing information, sales-related information, out-of-stock information, and promotion information corresponding to at least some of the identified items.

[0015] In one embodiment, the step of determining the target prediction model for analysis may include the step of identifying an item at risk of stock shortage among the plurality of items; and the step of determining a prediction model corresponding to the item at risk of stock shortage as the target prediction model for analysis.

[0016] In one embodiment, the step of identifying the item at risk of stock shortage may include the step of calculating individual item error information corresponding to each of the plurality of items designated for a plurality of periods; calculating the number of occurrences of underprediction errors among the plurality of periods based on the individual item error information; and identifying the item at risk of stock shortage based on the number of occurrences of underprediction errors and the total period corresponding to the entirety of the plurality of periods.

[0017] In one embodiment, the step of identifying the item at risk of stock shortage may include the step of calculating individual item error information corresponding to each of the plurality of items designated for a plurality of periods; the step of calculating the number of occurrences of underprediction errors exceeding a predetermined threshold error among the individual item error information corresponding to the plurality of periods; and the step of identifying the item at risk of stock shortage based on the number of occurrences of underprediction errors exceeding the threshold underprediction error.

[0018] In one embodiment, the step of identifying the items at risk of stock shortage may include the step of calculating a stock shortage risk ranking of each of the plurality of items based on the number of occurrences of underprediction errors corresponding to each of the plurality of items; and the step of identifying the items at risk of stock shortage based on the stock shortage risk ranking.

[0019] In one embodiment, the error cause analysis model includes a decision tree model including one or more rule nodes corresponding to each of the one or more decision rules, and the step of calculating the error cause information may include the steps of: determining, based on the one or more decision rules, a rule node corresponding to the expected sales amount calculated based on the target prediction model among the one or more rule nodes; identifying a result node corresponding to the target prediction model based on the determined rule node and the calculated expected sales amount; and calculating the error cause information corresponding to the result node.

[0020] In one embodiment, a level is assigned to each of the one or more rule nodes, and the decision tree model may include a rule node corresponding to a higher level and one or more rule nodes or one or more result nodes branched from the rule node corresponding to the higher level, wherein the one or more branched rule nodes or one or more result nodes correspond to a lower level adjacent to the higher level.

[0021] In one embodiment, one or more decision rules corresponding to the target prediction model for analysis may be determined based on pattern information identified based on at least some of the expected sales amount calculated based on the target prediction model for a predetermined past period, the actual sales amount corresponding to the predetermined past period, and error cause information calculated for the predetermined past period.

[0022] An electronic device according to one embodiment of the present disclosure comprises one or more processors, one or more memories storing instructions to be executed by the one or more processors, and when the instructions are executed by the one or more processors, the one or more processors may be configured to execute a method according to the present disclosure.

[0023] In a non-transitory computer-readable recording medium having recorded thereon instructions that, when executed by one or more processors according to one embodiment of the present disclosure, cause the one or more processors to perform operations, the instructions may be configured to cause the one or more processors to perform a method according to the present disclosure.

[0024] According to the present disclosure, the cause of errors in predicting sales opportunities for items in an e-commerce service can be more accurately determined.

[0025] According to the present disclosure, among various prediction models used to predict sales opportunities in e-commerce services, a target prediction model requiring analysis of error causes can be more accurately determined.

[0026] According to the present disclosure, the risk of stock shortages can be more accurately identified in e-commerce services.

[0027] The effects according to the technical idea of ​​the present disclosure are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the specification.

[0028] FIG. 1 illustrates an environment in which an electronic device according to one embodiment of the present disclosure can be applied.

[0029] FIG. 2 is a block diagram of an electronic device according to one embodiment of the present disclosure.

[0030] FIG. 3 is a flowchart illustrating a method for calculating error cause information according to one embodiment of the present disclosure.

[0031] FIG. 4 is a flowchart illustrating a method for determining a prediction model to be analyzed according to one embodiment of the present disclosure.

[0032] FIG. 5 is a diagram illustrating one or more prediction model consideration factors for determining a prediction model according to one embodiment of the present disclosure.

[0033] FIG. 6a and FIG. 6b are diagrams for more specifically explaining the method for calculating error information, error contribution, and prediction bias, which is described with reference to FIG. 3 and FIG. 4.

[0034] FIG. 7 is a flowchart illustrating a method for calculating error cause information according to one embodiment of the present disclosure.

[0035] Fig. 8 is a diagram for more specifically explaining the decision tree model described with reference to Fig. 7.

[0036] The various embodiments described in this disclosure are exemplified for the purpose of clearly explaining the technical concept of this disclosure and are not intended to be limited to specific embodiments. The technical concept of this disclosure includes various modifications, equivalents, alternatives, and embodiments selectively combined from all or part of the embodiments described in this disclosure. Furthermore, the scope of the technical concept of this disclosure is not limited to the various embodiments presented below or the specific descriptions thereof.

[0037] Terms used in this disclosure, including technical or scientific terms, unless otherwise defined, may have the meaning commonly understood by a person of ordinary skill in the art to which this disclosure belongs.

[0038] The expressions "includes," "may include," "comprises," "may have," "have," and "may have" used in this disclosure indicate the presence of a target feature (e.g., a function, operation, or component), but do not exclude the presence of other additional features. In other words, such expressions should be understood as open-ended terms that imply the possibility of including other embodiments.

[0039] The singular expressions used in this disclosure may include the plural meaning unless the context clearly indicates otherwise, and the same applies to the singular expressions set forth in the claims.

[0040] The expressions "first," "second," or "first", "second", etc. used in this document, unless the context indicates otherwise, are used to refer to multiple similar objects and to distinguish one object from another, and do not limit the order or importance among the objects.

[0041] As used herein, the expressions "A, B, and C", "A, B, or C", "A, B, and / or C", or "at least one of A, B, and C", "at least one of A, B, or C", "at least one of A, B, and / or C", etc., may refer to each of the listed items or to all possible combinations of the listed items. For example, "at least one of A or B" may refer to (1) at least one A, (2) at least one B, (3) at least one A and at least one B.

[0042] The expression "based on" as used in this disclosure is used to describe one or more factors that influence a decision, act of judgment, or action described in a phrase or sentence containing this expression, and this expression does not exclude additional factors that influence the decision, act of judgment, or action.

[0043] As used herein, the expression that a component (e.g., a first component) is “connected” or “connected” to another component (e.g., a second component) may mean that the component is directly connected or connected to the other component, as well as connected or connected via a new other component (e.g., a third component).

[0044] The expression "configured to" used in the present disclosure may have the meanings of "set to", "having the ability to", "modified to", "made to", "capable of", etc., depending on the context. This expression is not limited to the meaning of "specifically designed in hardware", and for example, a processor configured to perform a specific operation may mean a general purpose processor that can perform the specific operation by executing software, or a special purpose computer that is structured through programming to perform the specific operation.

[0045] Hereinafter, various embodiments of the present disclosure will be described with reference to the attached drawings. In the attached drawings and the description of the drawings, identical or substantially equivalent components may be assigned the same reference numerals. Furthermore, in the description of various embodiments below, duplicate descriptions of identical or corresponding components may be omitted, but this does not mean that such components are not included in the embodiments.

[0046] FIG. 1 illustrates an environment in which an electronic device according to an embodiment of the present disclosure can be applied. FIG. 1 illustrates an environment in which an electronic device according to an embodiment of the present disclosure can be applied. An electronic device (110) and a database (120) are connected via a network and can communicate with each other. Here, communication between the electronic device (110) and the database (120) may mean that the electronic device (110) communicates with a database management system (DBMS) that manages the database (120). In other words, the electronic device (110) can access the database (120) via a network.

[0047] The electronic device (110) may be a device for predicting sales opportunities for items in an e-commerce service and analyzing causes of errors in the prediction. The e-commerce service may include, for example, a sales platform that sells one or more items online. The electronic device (110) may be any server device, including a server device that manages the e-commerce service, or any electronic device, including an electronic device that manages inventory in the e-commerce service. As another example, the electronic device (110) may be implemented as a terminal capable of transmitting and receiving various information to and from a database (120) via a network. For example, the electronic device (110) may be one of a computer, a laptop, a portable communication terminal (such as a smartphone), a portable multimedia device, a wearable device, or an HMD. However, the type of the electronic device (110) is not limited thereto, and the electronic device (110) may be any device that includes an input / output interface capable of receiving information from a user or outputting information to a user, and that can communicate with the database (120) or other devices via a network. The electronic device (110) may utilize information stored in the database (120) to predict sales opportunities for an item and analyze causes of errors in the prediction. In one embodiment, the sales opportunity being predicted may include information regarding the potential likelihood of sales occurring for the item, such as at least some of the expected sales volume of the item, the expected loss sales volume of the item, and the expected loss sales volume of the item. In one embodiment, the error in the prediction may include at least some of the error that the predicted sales opportunity has with respect to actual sales, such as the error of the expected sales volume relative to the actual sales volume of the item, the error of the expected loss sales volume relative to the actual loss sales volume of the item, and the error of the actual loss sales volume relative to the actual expected loss sales volume of the item.

[0048] The electronic device (110) can provide various information regarding the prediction of sales opportunities and the analysis of error causes to the user, and can receive input from the user regarding the prediction of sales opportunities and the analysis of error causes. Specifically, the electronic device (110) can obtain a command regarding the prediction of sales opportunities from the user through an input / output interface, perform a prediction of sales opportunities in response to the obtained command, and output the results of the sales opportunity calculation to the user or store them in the electronic device (110), transmit them to another device, etc. Specifically, the electronic device (110) can obtain a command regarding the analysis of errors in the prediction of sales opportunities from the user through an input / output interface, perform an analysis of the causes of errors in the prediction of sales opportunities in response to the obtained command, and output the results of the error analysis to the user or store them in the electronic device (110), transmit them to another device, etc. The input obtained from the user can include various forms of input, such as a click using a mouse, a touch using a touchpad or a touch screen, voice recognition, other electronic input, or voice input using a microphone. The output of the electronic device (110) may include various forms of output, such as visual output through a display, projector, hologram, etc., or auditory output through a speaker, etc.

[0049] The database (120) can store information that serves as the basis for estimating sales opportunities and analyzing errors for items in e-commerce services. For example, the database (120) can store sales logs for each item handled in the e-commerce service. Furthermore, for example, the sales logs for each item in the database (120) can be updated in real time or repeatedly, either periodically or aperiodically. The electronic device (110) can obtain information regarding daily sales volume of any item through the database (120). For example, information regarding sales volume by period (e.g., date) of a specific item, all items handled in the e-commerce service, or items satisfying specific conditions among all items handled in the e-commerce service can be obtained. Furthermore, information regarding sales volume by time zone, day, week, month, quarter, year, or certain period can also be obtained for any item.

[0050] The network may serve to connect the electronic device (110) with the database (120) or other external devices. For example, the network may provide a connection path so that the electronic device (110) can be connected to the database (120) and transmit and receive packet data with the database (120). The network may be implemented as any type of wired or wireless network, such as a local area network (LAN), a wide area network (WAN), a mobile radio communication network, or Wibro (Wireless Broadband Internet).

[0051] Another example of an environment in which an electronic device according to another embodiment of the present disclosure can be applied, and which is not illustrated, is an electronic device (110) that may include a database. In this case, information stored in the database can be obtained without using the network illustrated in FIG. 1. In this case, the electronic device (110) may also include a database management system (DBMS) that manages the database.

[0052] FIG. 2 is a block diagram of an electronic device (200) according to one embodiment of the present disclosure. The electronic device (200) can process information regarding sales opportunity estimation and error cause analysis of an item. In one embodiment, the electronic device (200) may include one or more processors (210), one or more memories (220), and a communication interface (230) as components. In one embodiment, at least one of the components of the electronic device (200) may be omitted, or another component may be added to the electronic device (200). In one embodiment, additionally or alternatively, some of the components may be implemented in an integrated manner, or may be implemented as a single or multiple entities. In the present disclosure, one or more processors (210) may be referred to as a processor (210). The expression "processor (210)" may mean a set of one or more processors, unless the context clearly indicates otherwise. In the present disclosure, one or more memories (220) may be referred to as a memory (220). The expression "memory (220)" may mean a set of one or more memories, unless the context clearly indicates otherwise. In one embodiment, at least some of the components inside / outside the electronic device (200) may be connected to each other via a bus, a General Purpose Input / Output (GPIO), a Serial Peripheral Interface (SPI), or a Mobile Industry Processor Interface (MIPI), and may exchange information (data, signals, etc.).

[0053] The processor (210) may control at least one component of the electronic device (200) connected to the processor (210) by running software (e.g., commands, programs, etc.). In addition, the processor (210) may perform various operations such as calculations, processing, data generation, and processing related to the present disclosure. In addition, the processor (210) may load data, etc. from the memory (220) or store data in the memory (220). Furthermore, the processor (210) may transmit and receive various information with the database (120) through the communication interface (230). In one embodiment, the processor (210) may control the communication interface (230) to request various information related to sales opportunity estimation and error cause analysis from the database (120), and may receive various information related to sales opportunity estimation and error cause analysis from the database (120).

[0054] The memory (220) can store various information (data). The information stored in the memory (220) is information acquired, processed, or used by at least one component of the electronic device (200), and may include software (e.g., commands, programs, etc.). The memory (220) may include volatile and / or non-volatile memory. In the present disclosure, the commands or programs are software stored in the memory (220), and may include an operating system for controlling the resources of the electronic device (200), an application, and / or middleware for providing various functions to the application so that the application can utilize the resources of the electronic device (200). In one embodiment, the memory (220) may store commands that, when executed by the processor (210), cause the processor (210) to perform operations. The memory (220) may store at least a portion of information received from a database via the communication interface (230) and / or information transmitted to the database via the communication interface (230). Specifically, the memory (220) can store information regarding sales opportunity prediction and error cause analysis of an item and commands executed by the processor (210).

[0055] The communication interface (communication interface, 230) can perform wireless or wired communication between the electronic device (200) and a database or other external electronic device. For example, the communication interface (230) can perform wireless communication according to a method such as eMBB (enhanced Mobile Broadband), URLLC (Ultra Reliable Low-Latency Communications), MMTC (Massive Machine Type Communications), LTE (Long-Term Evolution), LTE-A (LTE Advance), NR (New Radio), UMTS (Universal Mobile Telecommunications System), GSM (Global System for Mobile communications), CDMA (Code Division Multiple Access), WCDMA (Wideband CDMA), WiBro (Wireless Broadband), WiFi (Wireless Fidelity), Bluetooth (Bluetooth), NFC (Near Field Communication), GPS (Global Positioning System), or GNSS (Global Navigation Satellite System). For example, the communication interface (230) may perform wired communication according to a method such as Universal Serial Bus (USB), High Definition Multimedia Interface (HDMI), Recommended Standard-232 (RS-232), or Plain Old Telephone Service (POTS). In one embodiment, the electronic device (200) may be implemented by being integrated with another device. In this case, the communication interface (230) may function as a connection circuit or interface connecting the electronic device (200) and the other device.

[0056] Hereinafter, the operations described as being performed by the electronic device in FIGS. 3 to 8 can be understood as being performed by the processor (210) of the electronic device (200) described in FIG. 2.

[0057] FIG. 3 is a flowchart illustrating a method for calculating error cause information according to one embodiment of the present disclosure.

[0058] Hereinafter, with reference to FIG. 3, the description will continue assuming that the operation of a method for calculating error cause information according to an embodiment of the present disclosure is performed by the electronic device (110) illustrated in FIG. 1. In addition, for convenience of explanation, the subject of the operation included in the method may be omitted, but unless otherwise indicated in the context, it may be interpreted that the operation is performed by the electronic device (110).

[0059] In operation S301, the electronic device (110) can determine a prediction model to be analyzed as an error cause among a plurality of prediction models.

[0060] In one embodiment, the predictive model may include one or more models that provide predictions regarding the sales opportunities of an item. The electronic device (110) may determine at least some of the multiple predictive models to predict sales opportunities. The electronic device (110) may predict sales opportunities of an item based on the determined predictive models.

[0061] A sales opportunity may include an expected sales volume of an item. The electronic device (110) may calculate the expected sales volume of an item based on a predictive model and characteristic information.

[0062] A predictive model may refer to multiple predictive models created, updated, or learned to reflect the expected sales pattern, sales trend, or user purchasing pattern of a specific item.

[0063] In one embodiment, the predictive model may include one or more models classified through methods such as learning, regression analysis, categorical analysis, or identified based on heuristic methods.

[0064] For example, the predictive model may include a predictive model (e.g., a promotional model) regarding the sales pattern of an item subject to a promotion. In an example of a predictive model for an item subject to a promotion, the electronic device (110) may determine that a condition for selecting a predictive model for the item subject to a promotion is satisfied if a promotion for the specific item is scheduled within a certain period of time (e.g., 35 days) from the prediction time. Accordingly, the electronic device (110) may predict that the sales pattern of the specific item will have a sales pattern corresponding to the promotional model, and may predict the sales volume of the item based on the sales pattern, i.e., based on the promotional model. The electronic device (110) may receive various information from the database (120) to determine whether the conditions for selecting the predictive model are satisfied.

[0065] In one embodiment, the predictive model may be determined based on predictive model consideration factors. The electronic device (110) may determine the predictive model to be used for sales prediction based on one or more predictive model consideration factors. The predictive model consideration factors may include at least some of item information, timing information, sales-related information, out-of-stock information, and promotional information.

[0066] The prediction model to be analyzed may be a prediction model whose error cause is to be analyzed.

[0067] In one embodiment, the electronic device (110) can determine a prediction model to be analyzed based on error information among multiple prediction models. A method for determining a prediction model to be analyzed based on error information is described in detail in FIG. 4.

[0068] In one embodiment, the electronic device (110) may determine the target prediction model for analysis based on the error contribution of each item. The error contribution may be the degree of contribution of the error in the estimated sales volume for each item to the error in the estimated sales volume for multiple items relative to the actual sales volume. A method for determining the target prediction model for analysis based on the error contribution is described in detail in FIG. 4.

[0069] In one embodiment, the electronic device (110) may determine a prediction model to be analyzed based on items at risk of stock shortage. Items at risk of stock shortage may refer to items at risk of stock shortage. For example, in the case of items with persistent underprediction errors, there is a possibility that stock may be insufficient compared to actual demand. The underprediction error may be an error due to underprediction and may refer to an error that occurs when the predicted sales volume is lower than the actual sales volume. The error corresponding to the underprediction may be calculated based on the weighted absolute average error between the actual sales volume and the predicted sales volume calculated when the underprediction occurs. The electronic device (110) may classify errors corresponding to overprediction and errors corresponding to underprediction among error information calculated for each of multiple items included in an item group. The error corresponding to the overprediction may refer to an error that occurs when the predicted sales volume is greater than the actual sales volume. The error corresponding to overprediction can be calculated based on the weighted absolute average error between actual and forecasted sales, calculated when overprediction occurs. A method for determining the target prediction model for analysis based on items at risk of stock shortages is described in detail in Figure 4.

[0070] In operation S302, the electronic device (110) can identify an error cause analysis model. The error cause analysis model can include multiple analysis models for analyzing the error causes of sales opportunities.

[0071] In one embodiment, the error source analysis model may include one or more decision rules. The decision rules may be conditions or rules for segmenting, classifying, or categorizing predicted data to derive error sources. The decision rules may correspond to categories of error sources, and the predicted data may include at least some of the predicted sales data, past sales data that serves as the basis for calculating the predicted sales data, and parameters and characteristic information of the prediction model. The decision rules are described in detail with reference to FIG. 8.

[0072] In one embodiment, an error source analysis model may be determined in response to the target prediction model being analyzed. Different error source analysis models may be determined for each target prediction model. One or more decision rules included in the error source analysis model may be determined in response to the target prediction model being analyzed.

[0073] In operation S303, the electronic device (110) may calculate error cause information based on the error cause analysis model. The electronic device (110) may calculate error cause information based on prediction data calculated based on the error cause analysis model and the analysis target prediction model. The error cause information may include information on one or more factors causing the error. For example, if the error cause is an input omission for sales information, the error cause information may be calculated as "sales information input omission." In this case, the sales information may be input data that serves as the basis for the prediction among the prediction data, and the error cause information corresponding to the sales information input omission may mean that an error in the prediction occurred due to an omission in the input data. The error cause information is described in detail in FIG. 8.

[0074] FIG. 4 is a flowchart illustrating a method for determining a prediction model to be analyzed according to one embodiment of the present disclosure.

[0075] Hereinafter, with reference to FIG. 4, the description will continue assuming that the operation of a method for determining a prediction model to be analyzed according to an embodiment of the present disclosure is performed by the electronic device (110) illustrated in FIG. 1. In addition, for convenience of explanation, the subject of the operation included in the method may be omitted, but unless otherwise indicated in the context, it may be interpreted that the operation is performed by the electronic device (110).

[0076] In operation S401, the electronic device (110) can calculate total error information. The total error information can be calculated based on the total expected sales amount and the total actual sales amount.

[0077] In one embodiment, the electronic device (110) can calculate a total expected sales volume corresponding to multiple items based on multiple prediction models. The multiple items may be items subject to sales volume prediction. In one embodiment, the electronic device (110) can determine multiple items subject to sales volume prediction.

[0078] The total expected sales volume may be the sum of the individual expected sales volumes calculated for each of the multiple items. The individual expected sales volume may be the expected sales volume calculated for each individual item. In one embodiment, the electronic device (110) may determine a prediction model for calculating the expected sales volume for each of the multiple items. A method for determining the prediction model is described in detail in FIG. 5. The electronic device (110) may calculate the individual expected sales volume for each item based on the prediction model corresponding to each item.

[0079] In one embodiment, the electronic device (110) may calculate expected sales corresponding to multiple time periods. For example, the electronic device (110) may calculate expected sales corresponding to multiple time periods for each item. The expected sales may be calculated for each time period included in the multiple time periods. The electronic device (110) may calculate expected sales corresponding to a unit time period. The unit time period may be, for example, one day. In this case, the multiple time periods may be multiple days. The electronic device (110) may calculate individual expected sales for each item corresponding to the multiple days, for each day.

[0080] In one embodiment, the electronic device (110) may calculate a total expected sales amount corresponding to a plurality of items. The electronic device (110) may calculate an individual expected sales amount corresponding to each of the plurality of items, and calculate a total expected sales amount based on the individual expected sales amounts.

[0081] In one embodiment, the electronic device (110) can calculate a total actual sales amount corresponding to a plurality of items based on a plurality of predictive models.

[0082] The total actual sales volume may be the sum of the individual actual sales volumes calculated for each of multiple items. The individual actual sales volume may be the actual sales volume calculated for each individual item.

[0083] In one embodiment, the electronic device (110) may calculate actual sales corresponding to multiple time periods. For example, the electronic device (110) may calculate actual sales corresponding to multiple time periods for each item. The actual sales may be calculated for each time period included in the multiple time periods. The electronic device (110) may calculate actual sales corresponding to a unit time period. The unit time period may be, for example, one day. In this case, the multiple time periods may be multiple dates. The electronic device (110) may calculate individual actual sales for each item, corresponding to each date. The electronic device (110) may determine multiple time periods for calculating individual actual sales corresponding to the time periods for which the calculated individual expected sales were calculated. For example, the electronic device (110) may calculate actual sales for the same time period as the time period for which the expected sales were predicted.

[0084] In one embodiment, the electronic device (110) may calculate a total actual sales amount corresponding to a plurality of items. The electronic device (110) may calculate an individual actual sales amount corresponding to each of the plurality of items, and calculate a total actual sales amount based on the individual actual sales amounts.

[0085] In one embodiment, the electronic device (110) may calculate total error information based on the total expected sales and the total actual sales. In one embodiment, the total error information may be calculated based on a Weighted Mean Absolute Percentage Error (WMAPE) value between the actual sales and the expected sales corresponding to the item group. The WMAPE may be calculated based on a weighted average of the absolute values ​​of the relative errors of the individual expected sales of each item relative to the individual actual sales, with the ratio of the individual actual sales of each item to the total actual sales corresponding to the entire plurality of items as a weight for each item. The WMAPE is described in detail in FIGS. 6A and 6B .

[0086] In operation S402, the electronic device (110) may calculate error contribution information corresponding to each of the plurality of items. In one embodiment, the electronic device (110) may calculate the error contribution of the first item based on first total error information corresponding to all of the plurality of items and second total error information corresponding to all of the remaining items in the item group excluding the first item. The error contribution may refer to the ratio of the error that a predicted value of a specific item (e.g., the first item) included in the plurality of items has with respect to an actual value to the total error of the entire plurality of items. The error contribution of the first item may be calculated based on the difference between the first total error information and the second total error information corresponding to all of the remaining items in the item group excluding the first item. The error contribution may have a negative value, and it may be understood that the smaller the error contribution of a specific item (e.g., the first item) has, the more accurately the prediction for the specific item (e.g., the first item) is made. In one embodiment, the error contribution information may be calculated for each item and for each period. For each period, the contribution of error in the expected sales of individual items to the expected sales of all items can be calculated.

[0087] In operation S403, the electronic device (110) can identify an error-causing target item. The electronic device (110) can identify an error-causing target item based on error contribution information.

[0088] In one embodiment, the electronic device (110) can identify an item with a large error contribution among all items. The electronic device (110) can calculate an error contribution corresponding to each individual item included in the all items. The electronic device (110) can identify an item as a source of error based on the ranking of the error contribution. For example, the electronic device (110) can determine an item with an error contribution corresponding to a top percentage (e.g., top 10%) among a plurality of items as a source of error. In one embodiment, the electronic device (110) can determine an item with an error contribution equal to or greater than a threshold value as a source of error.

[0089] In one embodiment, the electronic device (110) can identify target items that are the source of errors based on prediction bias information. The electronic device (110) can calculate prediction bias information corresponding to individual items or multiple items as a whole. The prediction bias information can include information regarding the tendency of errors exhibited by predicted results relative to actual values. For example, the prediction bias information can be calculated based on the error between the expected sales volume information and the actual sales volume information for an item.

[0090] In one embodiment, the individual prediction bias information corresponding to each item may be calculated based on a weighted absolute average error of the difference between the individual error information corresponding to overprediction and the individual error information corresponding to underprediction among the error information of the item calculated corresponding to each of a plurality of periods.

[0091] In one embodiment, prediction bias information may correspond to each item or each group of items. Even when predicting sales opportunities for the same item, the corresponding prediction models may differ for each period depending on the scenario selection conditions. For example, even for the same item, the satisfaction of the scenario selection conditions may vary depending on the period, resulting in different corresponding prediction models. The prediction bias information may be at least one of data reflecting the bias of the prediction values ​​corresponding to each item and the bias of the prediction values ​​corresponding to each prediction model.

[0092] In one embodiment, the electronic device (110) may calculate total prediction bias information corresponding to a group of items. The total prediction bias information may be calculated based on a weighted absolute mean error regarding the difference between error information corresponding to overprediction and error information corresponding to underprediction among error information corresponding to each item, calculated for all multiple items.

[0093] In one embodiment, the electronic device (110) can identify an item whose individual prediction bias value is greater than or equal to a specified threshold bias value as an error source target item based on an individual prediction bias corresponding to each of a plurality of items.

[0094] In one embodiment, the electronic device (110) can identify a group of items whose total prediction bias value is greater than or equal to a specified threshold bias value as a group of items causing errors based on the total prediction bias corresponding to each of a plurality of item groups.

[0095] In operation S404, the electronic device (110) may identify items at risk of running out of stock. Items at risk of running out of stock may refer to items at risk of running out of stock. For example, in the case of items with persistent underestimation errors, there is a possibility that inventory may be insufficient compared to actual demand. The underestimation error may be an error due to underestimation and may refer to an error that occurs when the predicted sales volume is lower than the actual sales volume. The error corresponding to the underestimation may be calculated based on the weighted absolute average error between the actual sales volume and the predicted sales volume calculated when the underestimation occurs. The electronic device (110) may classify errors corresponding to overestimation and errors corresponding to underestimation among error information calculated for each of a plurality of items included in an item group. The error corresponding to the overestimation may refer to an error that occurs when the predicted sales volume is greater than the actual sales volume. The error corresponding to overprediction can be calculated based on the weighted absolute mean error between actual sales and predicted sales when overprediction occurs.

[0096] In one embodiment, the electronic device (110) may calculate individual item error information for multiple time periods. Individual item error information refers to error information regarding the estimated sales volume calculated for an individual item relative to the actual sales volume of the individual item, and may include an underprediction error or an overprediction error.

[0097] In one embodiment, the electronic device (110) can calculate the number of occurrences of underprediction errors across multiple time periods. For example, if the multiple time periods are one week (7 days), one time period is set to one day (1 day), and individual item error information is calculated for each day, if underprediction errors occur for three days, the number of occurrences of underprediction errors can be '3'.

[0098] In one embodiment, the electronic device (110) may identify items at risk of stock shortage based on the number of occurrences of underprediction errors and the total number of periods. For example, if the ratio of the number of occurrences of underprediction errors for a specific individual item is greater than or equal to a threshold ratio for the total period corresponding to the total number of periods, the electronic device (110) may identify the specific individual item as an item at risk of stock shortage. In one embodiment, the electronic device (110) may calculate the number of occurrences of underprediction errors for each item and identify items at risk of stock shortage based on the ranking of the number of occurrences of underprediction errors for each item.

[0099] According to one embodiment, the electronic device (110) may calculate an underprediction error among individual item error information calculated for multiple periods, and may calculate the number of occurrences of underprediction errors exceeding a threshold error among the individual item errors corresponding to the multiple periods. The threshold error may be calculated as a pre-specified value or a certain percentage based on the error information. The electronic device (110) may identify items at risk of stock shortage based on the ranking of the number of occurrences of underprediction errors exceeding the threshold error.

[0100] In operation S405, the electronic device (110) can determine a prediction model to be analyzed based on at least some of the total error information, the error cause target items, and the items at risk of stock shortage.

[0101] In one embodiment, the electronic device (110) may set multiple item groups corresponding to each of multiple prediction models. The item groups corresponding to the prediction models may include multiple items for which a sales forecast has been performed based on the same prediction model. The electronic device (110) may calculate multiple total error information corresponding to each of the multiple item groups, and determine an item group with a higher error rate based on a comparison of the total error information. The item group with a higher error rate may be determined based on the ranking of the total error information. The electronic device (110) may determine the prediction model corresponding to the item group with the highest error rate as the prediction model to be analyzed.

[0102] In one embodiment, the electronic device (110) may calculate total error information corresponding to each prediction model for multiple items based on different prediction models. Based on a comparison of the total error information corresponding to each prediction model, the electronic device (110) may determine the prediction model with the highest error as the prediction model to be analyzed.

[0103] In one embodiment, the electronic device (110) may determine a prediction model corresponding to an error-causing target item as the analysis target prediction model. The electronic device (110) may determine the prediction model corresponding to the error-causing target item based on the prediction model consideration factors.

[0104] In one embodiment, the electronic device (110) may determine a prediction model corresponding to an item at risk of being out of stock as the target prediction model for analysis. The electronic device (110) may determine the prediction model corresponding to an item at risk of being out of stock based on the prediction model consideration factors.

[0105] FIG. 5 is a diagram illustrating one or more prediction model consideration factors for determining a prediction model according to one embodiment of the present disclosure.

[0106] Referring to Table (400) regarding the predictive model consideration factors of FIG. 5, the predictive model consideration factors may include at least some of item information, timing information, sales-related information, out-of-stock information, and promotional information. The predictive model consideration factors may include one or more factors that may influence the sales pattern of an item.

[0107] In one embodiment, item information may include various information about the item. Item information may include the item's category and at least some of the item's attributes. The item's category may be classified into one or more categories (e.g., a first category, a second category, a third category, a fourth category, and a fifth category).

[0108] In one embodiment, the attributes of an item may include at least some of order status, stocking period, selling period, seasonality, and holiday characteristics.

[0109] In one embodiment, the order status may include various information regarding the order of an item. An order for an item can be understood as a process for purchasing an item from a supplier and storing it in stock for sale to customers in an e-commerce service. The order status may include at least some of the following: whether the item has been ordered, the order date, the order quantity, whether the order has been completed, whether the order has been canceled, and whether the item has been received after the order has been completed.

[0110] In one embodiment, the stocking period may include information regarding the period during which an item for sale was received in a storage facility, warehouse, fulfillment center, etc. For example, the stocking period may correspond to the period from the first stocking to the last stocking for the same item.

[0111] In one embodiment, the sale period may correspond to the period from the time a particular item was first sold to the time it was last sold.

[0112] In one embodiment, seasonality may correspond to the characteristic that an item's sales volume is affected by the season. The seasonality of an item may be determined based on a correlation between the item and temperature, calculated based on historical sales data and temperature data for each period. Seasonality may include, for example, at least some information about whether the item's sales volume is affected by the season (whether seasonality exists) and the season that affects the item's sales volume. For example, if an item has the characteristic of high sales during the "summer" period and low sales during other seasons, the seasonality of the item may be "summer." In this case, the "season" in "seasonality" does not necessarily refer to the four meteorological seasons, such as spring, summer, fall, and winter. Seasonality may refer to a characteristic that fluctuates over a period of time, such as the rainy season, the beginning of the year, and the end of the year.

[0113] In one embodiment, a holiday attribute may refer to an attribute that affects an item's sales volume on holidays. The holiday attribute may include at least some information about whether the holiday attribute exists and about specific holidays that affect the item's sales volume. For example, if an item's sales surge only during the Chuseok holiday period, the holiday attribute for that item may be "Chuseok."

[0114] In one embodiment, the timing information may include actual sales data that can be used to predict sales opportunities, or information indicating attributes of a time period corresponding to the period for which sales opportunities are to be predicted. The timing information may include at least some of the following: week of the year, week of the month, season, day of the week, date of the month, holiday information, seasonal information, and promotional schedule information. In one embodiment, the timing information may correspond to a specific date or period, and multiple timing information may correspond to a specific date or period.

[0115] Week of year can refer to the week number of a specific time period within the same year. Week of month can refer to the week number of a specific time period within the same month.

[0116] In one embodiment, sales-related information may include at least some of sales volume information, price information, sales summary information, and restock notification information. The sales summary information may include key information about sales for a specific item or over a specific period. For example, it may include at least some information about average sales, top-selling items, top-selling items, and inventory levels. Restock notification information may include information about whether a specific item is available for restock.

[0117] In one embodiment, the out-of-stock information may include the out-of-stock period. The out-of-stock period may refer to the period during which a specific item remains out-of-stock due to lack of stock. The out-of-stock period may refer to the percentage of time the item remains out-of-stock within a reference period. For example, if one day is used as the reference period and the item remains out-of-stock for six hours, the out-of-stock period may be expressed as a percentage, such as 0.25 or 25%, representing the percentage of six hours in a day (24 hours).

[0118] In one embodiment, promotion information may include at least some of the promotion type and promotion details. Promotion types may include various types, such as promotions that discount an item by a certain amount, promotions that discount an item by a certain percentage, promotions that issue coupons, promotions that utilize rewards, seasonal promotions, regular promotions, and promotions for preferred customers. Promotion details may include various information about the promotion, such as the promotion period, the promotion discount rate, the category of items eligible for the promotion, and the category of customers eligible for the promotion.

[0119] In one embodiment, the electronic device (110) may determine whether a prediction model selection condition is satisfied based on one or more prediction model consideration factors. The prediction model may be determined based on one or more prediction model consideration factors corresponding to the item being predicted and the period being predicted.

[0120] In one embodiment, the electronic device (110) may determine at least some of a plurality of predictive models for calculating sales opportunities as predictive models based on predictive model consideration factors identified corresponding to items and periods for which sales opportunities are to be calculated.

[0121] In one embodiment, the plurality of prediction models may include a first prediction model, a second prediction model, a third prediction model, a fourth prediction model, a fifth prediction model, a sixth prediction model, a seventh prediction model, and a base prediction model, but the prediction models according to the present disclosure are not limited thereto.

[0122] In one embodiment, the first prediction model may include a prediction model corresponding to a new item. A new item may refer to a newly released item or an item that was not sold in the past but has recently started selling. Items corresponding to new items may have similar sales patterns, and such patterns may be reflected in the prediction model corresponding to the new item. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether the first prediction model condition corresponding to the first prediction model is satisfied. Whether the first prediction model condition is satisfied may be determined based on the creation date of the item information. For example, if the prediction target period and item for which a sales opportunity is to be predicted are within a specified period (e.g., 30 days) from the creation date of the item information for the corresponding item, the electronic device (110) may determine that the first prediction model condition is satisfied. To determine whether the first prediction model condition is satisfied, the electronic device (110) may receive period information and item information among the prediction model consideration factors from the database.

[0123] In one embodiment, the electronic device (110) may further check sales-related information to determine whether the first prediction model condition is satisfied. The first prediction model condition may be determined based on at least some of the following: whether the initial sales occurred within the specified period from the creation date of the item information, in addition to whether the prediction target time (e.g., the current point in time or the prediction target date) is within a specified period from the creation date of the item information; and whether restocking notification information exists within the specified period. For example, if item information is created as a new item but a specified period has passed since then, it may not be appropriate to predict sales opportunities for the item based on the sales pattern of the new item. Similarly, if no sales occurred until the specified period has passed, it may not be appropriate to predict sales for the item based on the sales pattern corresponding to the new item.

[0124] In one embodiment, the second prediction model may include a prediction model corresponding to aging items. Aging items may include items that are sensitive to trends or the latest technology, items with a rapid new product release cycle, or items that tend to see a decline in sales as inventory ages. For example, smartphones have a rapid new product release cycle and are sensitive to the latest technology, so sales may decline rapidly as inventory ages. Items corresponding to aging items may have similar sales patterns, and such patterns may be reflected in the prediction model corresponding to aging items. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether the second prediction model conditions corresponding to the second prediction model are satisfied. Whether the second prediction model conditions are satisfied may be determined based on at least some of the order status, sales volume information, and the delivery period. For example, whether the second prediction model conditions are satisfied may be determined based on whether sales volume has decreased by a threshold value after a certain period of time based on past sales volume information. As another example, the electronic device (110) may check the order status and determine whether the second prediction model condition is satisfied based on whether an order exists after a certain period of time. As another example, the electronic device (110) may determine that the second prediction model condition is satisfied if the sales volume decreases by a threshold value after a certain period of time and there are no orders after the certain period of time. In order to determine whether the second prediction model condition is satisfied, the electronic device (110) may receive order information and sales volume information from the database, among the prediction model consideration factors.

[0125] In one embodiment, the third prediction model may include a prediction model corresponding to slow-moving items. A slow-moving item may refer to an item with a low turnover rate, i.e., an item that tends to remain in inventory due to slow sales or low demand. Items corresponding to slow-moving items may have similar sales patterns, and such patterns may be reflected in the prediction model corresponding to slow-moving items. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether a third prediction model condition corresponding to the third prediction model is satisfied. Whether the third prediction model condition is satisfied may be determined based on price information, sales volume information, and / or item attributes (e.g., stocking period). For example, the third prediction model condition may include whether the price of the item is above a specified threshold price. If the price of the item is high, demand may be low, which may lead to a lower turnover rate. As another example, the third prediction model condition may include whether the sales volume of the item has been below a threshold sales volume during a specified past period. For example, based on the period to be predicted, if sales less than the critical sales amount occurred during a specified period in the past (e.g., 3 weeks) or the number of days (days) on which sales occurred is less than the critical number of days, the electronic device (110) may determine that the third prediction model condition is satisfied. As another example, whether the third prediction model condition is satisfied may be determined based on the period of time during which the item was received. Whether the third prediction model condition is satisfied may be determined based on whether the first time of receipt of the item is after a specified period (e.g., 60 days) from the predicted date. In order to determine whether the third prediction model condition is satisfied, the electronic device (110) may receive period information and item information, among the prediction model consideration factors, from the database.

[0126] In one embodiment, the fourth prediction model may include a prediction model corresponding to a promotional item. A promotional item may refer to an item scheduled for promotion. Items corresponding to promotional items may have similar sales patterns, and a prediction model corresponding to a promotional item may be created, updated, or improved to reflect such patterns. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether a fourth prediction model condition corresponding to the fourth prediction model is satisfied. Whether the fourth prediction model condition is satisfied may be determined based on whether a promotion is scheduled within a specified period of time after the prediction date from the prediction date. For example, based on the target time and item for predicting a sales opportunity, if promotion schedule information exists within a specified period of time (e.g., 35 days) after the target time of the prediction, and if the target time of the prediction is during a promotion period, the fourth prediction model condition may be determined to be satisfied. The electronic device (110) may determine whether the fourth prediction model condition is satisfied based on at least some of the item attributes, promotion schedule information, and promotion information among the prediction model consideration factors. The electronic device (110) can receive item attributes, promotion schedule information, and promotion information among the prediction model consideration factors from the database (120) to determine whether the fourth prediction model condition is satisfied.

[0127] In one embodiment, the fifth prediction model may include a prediction model corresponding to an item having a holiday characteristic. An item having a holiday characteristic may refer to an item whose sales volume fluctuates significantly based on whether it falls on a specific holiday. For example, for an item that has high sales only during the Chuseok holiday period and low sales during other periods, the fifth prediction model corresponding to the holiday characteristic may be applied. Items corresponding to the holiday characteristic may have similar sales patterns, and such patterns may be reflected in the fifth prediction model corresponding to the item having the holiday characteristic. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether the fifth prediction model condition corresponding to the fifth prediction model is satisfied. Whether the fifth prediction model condition is satisfied may be determined based on whether the item has a holiday characteristic and whether a holiday is scheduled within a specified period from the prediction date to the prediction date. For example, based on the target time and the attributes of the item for predicting a sales opportunity, if the item has a holiday characteristic and holiday information exists within a specified period (e.g., 35 days) after the target time for predicting, the fifth prediction model condition may be determined to be satisfied. In one embodiment, the electronic device (110) may determine whether the fifth prediction model condition is satisfied based on whether the holiday information and the holiday corresponding to the holiday characteristic are corresponding types of holidays. For example, among the prediction model consideration elements, the holiday characteristic and the holiday information may each include information about the type of the corresponding holiday (e.g., Lunar New Year, Chuseok, Christmas). The electronic device (110) may determine whether the fifth prediction model condition is satisfied based on at least some of the holiday characteristic and the holiday information among the prediction model consideration elements.The electronic device (110) can receive holiday characteristics and holiday information from the database among the prediction model consideration factors to determine whether the fifth prediction model condition is satisfied.

[0128] In one embodiment, the sixth prediction model may include a prediction model corresponding to a seasonal item. A seasonal item may refer to an item whose sales volume fluctuates significantly depending on whether it falls in a specific season. For example, for an item that has high sales volume only during the summer and low sales volume during other periods, the sixth prediction model corresponding to the seasonal item may be applied. Items corresponding to seasonal items may have similar sales patterns, and such patterns may be reflected in the sixth prediction model corresponding to the seasonal item. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether the sixth prediction model condition corresponding to the sixth prediction model is satisfied. Whether the sixth prediction model condition is satisfied may be determined based on whether the item is seasonal and whether a specific season is expected within a specified period from the prediction date to the prediction date. For example, based on the time period and the attributes of the item to be predicted for a sales opportunity, if the item has seasonality and seasonal information exists within a specified period (e.g., 35 days) after the time period to be predicted, it may be determined that the sixth prediction model condition is satisfied. In one embodiment, the electronic device (110) may determine whether the sixth prediction model condition is satisfied based on whether the seasonal information and the season corresponding to the seasonality are corresponding types of seasons. For example, among the prediction model consideration factors, seasonality and season information may each include information about the type of the corresponding season (e.g., spring, summer, fall, winter, rainy season, dry season, rainy season, specific season, white night, polar night, weather). The electronic device (110) may determine whether the sixth prediction model condition is satisfied based on at least some of the seasonality and season information among the prediction model consideration factors.The electronic device (110) can receive seasonality and seasonal information from the database among the prediction model consideration factors to determine whether the sixth prediction model condition is satisfied.

[0129] In one embodiment, the seventh prediction model may include a prediction model corresponding to an item having both a holiday characteristic and a seasonal characteristic. Items corresponding to items having both a holiday characteristic and a seasonal characteristic may have similar sales patterns, and such patterns may be reflected in the seventh prediction model corresponding to items having both a holiday characteristic and a seasonal characteristic. The electronic device (110) may determine that the seventh prediction model condition is satisfied when a specific season and a specific holiday exist simultaneously within a specified period (e.g., 35 days) from the period being predicted, and the item's attributes include both seasonal and holiday characteristics. The electronic device (110) may check at least some information among the prediction model consideration factors to determine whether the seventh prediction model condition corresponding to the fifth prediction model is satisfied. Whether the seventh prediction model condition is satisfied may be determined based on whether the item has a holiday characteristic, whether the item has seasonality, and whether a specific holiday and a specific season are scheduled simultaneously within a specified period from a prediction date to a prediction date. For example, based on the time period and the attributes of the item to be predicted for a sales opportunity, if the item has seasonality and holiday characteristics, and seasonal information and holiday information exist within a specified period (e.g., 35 days) after the time period to be predicted, it may be determined that the seventh prediction model condition is satisfied. The electronic device (110) may determine whether the seventh prediction model condition is satisfied based on at least some of seasonality, seasonal information, holiday characteristics, and holiday information among the prediction model consideration factors. In order to determine whether the seventh prediction model condition is satisfied, the electronic device (110) may receive seasonality, seasonal information, holiday characteristics, and holiday information among the prediction model consideration factors from a database.

[0130] The basic prediction model may refer to a model selected for predicting sales volume in response to items that do not correspond to the first to seventh prediction models, i.e., items that include general sales patterns.

[0131] In one embodiment, the first prediction model may include a weighted moving average model algorithm. The electronic device (110) may select the first prediction model and determine one or more characteristic pieces of information to be applied to the first prediction model. The one or more characteristic pieces of information corresponding to the first prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the first prediction model may include, for example, at least a portion of item information, time information, sales information, and out-of-stock information. In one embodiment, the electronic device (110) may generate, update, or learn the first prediction model by updating one or more parameters to be applied to the algorithm of the first prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (the first prediction model). The electronic device (110) may calculate the expected sales amount of the item based on the first prediction model and the one or more characteristic pieces of information.

[0132] In one embodiment, the second prediction model may include a Moving Average model algorithm. When the second prediction model is selected, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the second prediction model. The one or more characteristic pieces of information corresponding to the second prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the second prediction model may include, for example, at least a portion of order status, item information, timing information, sales information, and out-of-stock information. In one embodiment, the electronic device (110) may generate, update, or learn the second prediction model by updating one or more parameters to be applied to the algorithm of the second prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (the second prediction model). The electronic device (110) may calculate the expected sales amount of the item based on the second prediction model and the one or more characteristic pieces of information.

[0133] In one embodiment, the third prediction model may include a Moving Average model algorithm. If the electronic device (110) corresponds to the third prediction model, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the third prediction model. The one or more characteristic pieces of information corresponding to the third prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the third prediction model may include, for example, at least a portion of a stocking period, item information, timing information, sales volume information, and out-of-stock information. In one embodiment, the electronic device (110) may generate, update, or learn the third prediction model by updating one or more parameters to be applied to the algorithm of the third prediction model based on existing prediction data and past sales volume information regarding one or more items corresponding to the same prediction model (the third prediction model). The electronic device (110) may calculate the expected sales volume of the item based on the third prediction model and the one or more characteristic pieces of information.

[0134] In one embodiment, the fourth predictive model may include at least a portion of a Light Gradient-Boosting Machine (Light GBM) algorithm and an Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) algorithm. When the electronic device (110) corresponds to the fourth predictive model, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the fourth predictive model. The one or more characteristic pieces of information corresponding to the fourth predictive model may include at least a portion of predictive model consideration factors. The one or more characteristic pieces of information to be applied to the fourth predictive model may include at least a portion of item characteristics, timing information, sales information, out-of-stock information, and promotion information, for example. In one embodiment, the electronic device (110) may generate, update, or learn the fourth predictive model by updating one or more parameters to be applied to the algorithm of the fourth predictive model based on existing prediction data and past sales information regarding one or more items corresponding to the same predictive model (the fourth predictive model). The electronic device (110) can calculate the expected sales amount of an item based on the fourth prediction model and one or more characteristic pieces of information.

[0135] In one embodiment, the fifth prediction model may include a Holt-Winters algorithm. The electronic device (110) may determine one or more characteristic pieces of information to be applied to the fifth prediction model, if the fifth prediction model corresponds to the fifth prediction model. The one or more characteristic pieces of information corresponding to the fifth prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the fifth prediction model may include, for example, at least a portion of item attributes, timing information, timing information, sales information, and out-of-stock information. In one embodiment, the electronic device (110) may generate, update, or learn the fifth prediction model by updating one or more parameters to be applied to the algorithm of the fifth prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (the fifth prediction model). The electronic device (110) may calculate the expected sales amount of the item based on the fifth prediction model and the one or more characteristic pieces of information.

[0136] In one embodiment, the sixth prediction model may include at least a portion of the Light Gradient-Boosting Machine (Light GBM) algorithm and the Holt-Winters algorithm. When the electronic device (110) corresponds to the sixth prediction model, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the sixth prediction model. The one or more characteristic pieces of information corresponding to the sixth prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the sixth prediction model may include, for example, at least a portion of item attributes, timing information, sales information, out-of-stock information, sales summary information, seasonal information, and price information. In one embodiment, the electronic device (110) may generate, update, or learn the sixth prediction model by updating one or more parameters to be applied to the algorithm of the sixth prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (the sixth prediction model). The electronic device (110) may calculate the expected sales of the item based on the sixth prediction model and the one or more characteristic pieces of information.

[0137] In one embodiment, the seventh prediction model may include at least a portion of the Holt-Winters algorithm. If the electronic device (110) corresponds to the seventh prediction model, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the seventh prediction model. The one or more characteristic pieces of information corresponding to the seventh prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the seventh prediction model may include, for example, at least a portion of item attributes, time information, sales information, out-of-stock information, sales summary information, holiday information, seasonal information, and price information. In one embodiment, the electronic device (110) may generate, update, or learn the seventh prediction model by updating one or more parameters to be applied to the algorithm of the seventh prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (the seventh prediction model). The electronic device (110) may calculate the expected sales of the item based on the seventh prediction model and the one or more characteristic pieces of information.

[0138] In one embodiment, the basic prediction model may include a Light GBM (Light Gradient-Boosting Machine) algorithm. When the electronic device (110) corresponds to a general sales prediction model, the electronic device (110) may determine one or more characteristic pieces of information to be applied to the basic prediction model. The one or more characteristic pieces of information corresponding to the basic prediction model may include at least a portion of the prediction model consideration factors. The one or more characteristic pieces of information to be applied to the basic prediction model may include, for example, at least a portion of item attributes, timing information, sales-related information, and out-of-stock information. Specifically, the characteristic pieces of information corresponding to the basic prediction model may include at least a portion of order status, stocking period, and sales period among item attributes. The characteristic pieces of information corresponding to the basic prediction model may include at least a portion of week within the year, week within the month, day of the week, and day within the month among period information. In one embodiment, the electronic device (110) may generate, update, or learn the basic prediction model by updating one or more parameters to be applied to the algorithm of the basic prediction model based on existing prediction data and past sales information regarding one or more items corresponding to the same prediction model (basic sales prediction model). The electronic device (110) can calculate the expected sales amount of an item based on a basic prediction model and one or more characteristic pieces of information.

[0139] According to one embodiment, the electronic device (110) can identify an error-causing target item and determine a prediction model that serves as the basis for calculating sales corresponding to the identified error-causing target item based on prediction model selection conditions. The electronic device (110) can then determine the determined prediction model as the analysis target prediction model.

[0140] According to one embodiment, the electronic device (110) can identify items at risk of being out of stock and determine a predictive model that serves as the basis for estimating sales volume corresponding to the identified items at risk of being out of stock, based on predictive model selection conditions. The electronic device (110) can then determine the determined predictive model as the target predictive model for analysis.

[0141] FIG. 6a and FIG. 6b are diagrams for more specifically explaining the method for calculating error information, error contribution, and prediction bias, which is described with reference to FIG. 3 and FIG. 4.

[0142] In FIG. 6a, a method for calculating error information, error contribution, and prediction bias information is specifically described based on an example table (601) regarding sales opportunities, which shows expected sales and actual sales corresponding to a plurality of dates (20230101 and 20230102) as examples for a plurality of periods, and an item group including a first item, a second item, and a third item as examples for a plurality of items.

[0143] In Fig. 6b, a method for calculating sales opportunities and corrected sales opportunities based on an error calculation example table (602) showing error information and prediction bias information calculated based on the sales opportunity example table (601) of Fig. 6a is specifically described.

[0144] Referring to FIG. 6a, the expected sales volume of the sales opportunity example table (601) can be calculated for each item and each period (by date).

[0145] Referring to FIG. 6A, error information can be calculated based on mathematical expression 1 (610), an error corresponding to overprediction can be calculated based on mathematical expression 2 (620), and an error corresponding to underprediction can be calculated based on mathematical expression 3 (630). In addition, prediction bias information can be calculated based on mathematical expression 4 (640). At this time, "i" may be an index corresponding to each item, and "n" may be the final value of the item index. For example, in the example of FIG. 6A, the values ​​"1", "2", and "3" may be assigned to the index values ​​("i") of the first item, the second item, and the third item, respectively, and the value "3" may be assigned to "n" corresponding to the final value of the item index. At this time, "A i " may be the actual sales corresponding to the item index "i", and "F i " may be the expected sales corresponding to item index "i".

[0146] Referring to FIG. 6B, the calculation example table (602) may include results calculated for all items included in the item group among the results calculated based on the numerical values ​​included in the sales opportunity example table (601) of FIG. 6A and mathematical equations 1 (610), 2 (620), 3 (630), and 4 (640). The total error information (77%) for the entire item group corresponding to the date "20230101" may be calculated by substituting the expected sales amount and actual sales amount corresponding to "20230101" of the first item, the second item, and the third item, respectively, into mathematical equation 1 (610). Through the same method, based on mathematical expressions 1 (610), 2 (620), 3 (630), and 4 (640), the total error information (45%) corresponding to the date "20230102", the total overprediction error (73%) corresponding to the date "20230101", the total underprediction error (5%) corresponding to the date "20230101", and the total prediction bias information (68%) corresponding to the date "20230101" can be calculated, and the total error information (80%) corresponding to all items except the first item corresponding to the date "20230101" can be calculated. At this time, for the total error information corresponding to all items except the first item, only "2" and "3" excluding "1" corresponding to the first item can be substituted into the item index "i" of mathematical expression 3 (630).

[0147] Referring to FIG. 6b, the error contribution (-3%) of the first item corresponding to the date “20230101” can be calculated as the difference between the total error (67%) of all items excluding the first item corresponding to “20230101” and the total error (80%) of all items excluding the first item.

[0148] Referring to FIG. 6B, the prediction bias information (-11%) corresponding to the first item in the period "20230101" to "20230102" can be calculated by applying different indices for each date to the first item in mathematical expression 4 (640) described with reference to FIG. 6A, and inputting the expected sales amount and the actual sales amount. That is, "1" can be input to the index "i" for the expected sales amount and the actual sales amount of the first item in "20230101", and "2" can be input to the index "i" for the expected sales amount and the actual sales amount of the first item in "20230102".

[0149] FIG. 7 is a flowchart illustrating a method for calculating error cause information according to one embodiment of the present disclosure.

[0150] In operation S701, the electronic device (110) may determine a rule node corresponding to the expected sales volume based on a decision rule. The rule node may be a node related to one or more decision rules included in the error cause analysis model.

[0151] The projected sales volume may include predicted data that is segmented, classified, or categorized by decision rules. The predicted data may include at least some of the projected sales volume data, past sales volume data that serves as the basis for calculating the projected sales volume data, and parameters and characteristic information of the prediction model.

[0152] In one embodiment, the error source analysis model may include a decision tree model including one or more rule nodes.

[0153] One or more rule nodes may correspond to decision rules. A decision rule may be a condition or rule for segmenting, classifying, or categorizing prediction data to derive error sources. One or more decision rules may be determined in response to the prediction model being analyzed.

[0154] In one embodiment, the decision rules corresponding to each predictive model may include hypothesis-based rules learned based on patterns of past predicted sales, past actual sales, and errors therebetween for each predictive model. For example, the electronic device (110) may determine decision rules corresponding to the predictive models based on rule-based learning. One or more decision rules may be determined based on pattern information identified based on at least a portion of the predicted sales produced based on the target predictive model for a predetermined past period, the actual sales corresponding to the predetermined past period, and error cause information produced during the predetermined past period.

[0155] In one embodiment, a decision tree model may include a rule node corresponding to a higher level, and one or more rule nodes or one or more result nodes branching from the rule node at the higher level. One or more rule nodes or one or more result nodes branching from the rule node at the higher level may correspond to lower levels adjacent to the higher level. For example, one or more rule nodes corresponding to a third level may branch from a rule node at the second level.

[0156] In operation S702, the electronic device (110) can identify a result node based on the determined rule node and the expected sales volume. The result node can correspond to error cause information. The result node can include adjacent lower-level nodes corresponding to some rule nodes among one or more rule nodes. For example, the result node can refer to a node corresponding to the final level, i.e., a node branching from an upper level adjacent to the final level. The electronic device (110) can identify a result node from the rule node corresponding to the expected sales volume, again based on the expected sales volume.

[0157] In operation S703, the electronic device (110) may calculate error cause information based on the determined result node. In one embodiment, the error cause information may have a corresponding relationship with each result node. The electronic device (110) may output error cause information corresponding to the identified result node.

[0158] Fig. 8 is a diagram for more specifically explaining the decision tree model described with reference to Fig. 7.

[0159] Referring to FIG. 8, FIG. 8 may be a drawing illustrating an example (800) of a decision tree model included in an error cause analysis model.

[0160] The decision tree model may correspond to the target prediction model (801) to be analyzed, which is the subject of error analysis. The decision tree model may include one or more rule nodes and result nodes corresponding to the target prediction model (801). For example, in one example (800), a promotion model may be selected as the target prediction model (801) to be analyzed, and the decision tree model may include one or more rule nodes and result nodes corresponding to the promotion model.

[0161] A decision tree model may include one or more levels, and each level may include one or more rule nodes or result nodes. For example, level 1 (810) may include a plurality of rule nodes corresponding to the target prediction model (801), namely, an input-related rule node (811) and a model-related rule node (812). The input-related rule node (811) may correspond to a decision rule regarding whether the cause of an error determined based on prediction data occurred in relation to the input of data. The model-related rule node (812) may correspond to a decision rule regarding whether the cause of an error determined based on prediction data occurred in relation to any one of the application of the prediction model, the processing of the model, and the post-processing of the prediction model. The electronic device (110) may determine whether the input data is classified into any one of the input-related rule node (811) and the model-related rule node (812) based on at least some of the predicted sales data, the past sales data that serves as the basis for calculating the predicted sales data, and the parameters and characteristic information of the prediction model, which are included in the predicted data. Predictive data may include forecast data regarding expected sales volume calculated by date and by item.

[0162] In one embodiment, in response to level 2 (820), an operation-related rule node and an ETL rule node (821) branched from an input-related rule node (811) may be included. The operation-related rule node may correspond to a decision rule regarding whether an error cause occurred in an input related to the operation of a prediction model among error causes related to the input of prediction data. The ETL (Extract, Transform, Load) rule node (821) may correspond to a decision rule regarding whether an error cause occurred based on any one of extraction, transformation, and loading of prediction data. The electronic device (110) may identify any one of the operation-related rule node and the ETL rule node (821) to be classified from the input-related rule node (811) based on the prediction data.

[0163] Likewise, at level 1 (810), if the input prediction data is classified as a model-related rule node (812), the electronic device (110) can identify one rule node among the model applicability rule node, the model processing rule node, and the post-processing rule node to which the prediction data will be classified at level 2 (820). The model applicability rule node can be identified when there is an error in the characteristic information applicable to the prediction model (e.g., the item to be applied), the model processing rule node can be identified when an error occurs in the processing of the input data in the algorithm of the prediction model, and the post-processing rule node can be identified when an error occurs in the post-processing of the prediction value of the prediction model, such as the correction of the prediction value or the selection of an appropriate prediction model.

[0164] In one embodiment, the decision tree model may include an input missing rule node (831), an input delayed rule node (832), and an input error rule node (833) branching from an ETL rule node (821), corresponding to level 3 (830).

[0165] In addition, the decision tree model can branch from the rule nodes regarding input omission, input delay, and input error at level 4 (840) into multiple rule nodes each corresponding to whether the input corresponds to any one of sales information, out-of-stock information, price information, promotion information, sales information, and item information among the characteristic information input into the prediction model. For example, if the electronic device (110) identifies the input omission rule node (831) based on the prediction data at level 3 (830), it can identify the sales information rule node (841) among the sales information, out-of-stock information, price information, promotion information, sales information, and item information based on whether any data is missing from the prediction data among the sales information, out-of-stock information, price information, promotion information, sales information, and item information.

[0166] A node of level 5 (850) branching from a decision rule node of level 4 (840) may be a result node, and the result node may correspond to error cause information. For example, a sales information rule node (841) of level 4 may branch from an input omission rule node (831) of level 3 (830), and the corresponding result node of level 5 (850) may be a sales information input omission (851) node. When a result node is identified, the electronic device (110) may produce error cause information corresponding to the result node.

[0167] In the flowcharts according to the present disclosure, each step of the method or algorithm is described in a sequential order. However, the steps may be performed in any order that can be arbitrarily combined, in addition to being performed sequentially. The description of the flowcharts or flowcharts of the present disclosure does not exclude changes or modifications to the method or algorithm, and does not imply that any step is essential or desirable. In one embodiment, at least some of the steps may be performed in parallel, iteratively, or heuristically. In another embodiment, at least some of the steps may be omitted, or other steps may be added.

[0168] Various embodiments according to the present disclosure may be implemented as software on a machine-readable storage medium. The software may be software for implementing various embodiments described in the present disclosure. The software may be inferred from various embodiments described in the present disclosure by programmers skilled in the art to which the present disclosure pertains. For example, the software may be a program including machine-readable instructions (e.g., instructions, codes, or code segments). The device may be a device capable of operating according to instructions called from a storage medium, such as a computer. In one embodiment, the device may be a computing device according to various embodiments described in the present disclosure. In one embodiment, the processor of the device may execute the called instructions, causing components of the device to perform functions corresponding to the instructions. The storage medium may refer to any type of recording medium that stores data and can be read by the device. The storage medium may include, for example, ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, etc. In one embodiment, the storage medium may be implemented in a distributed form, such as in a network-connected computer system. In this case, the software may be distributed, stored, and executed in the computer system. In another embodiment, the storage medium may be a non-transitory storage medium. A non-transitory storage medium refers to a tangible medium that exists regardless of whether data is stored semi-permanently or temporarily, and does not include signals that are transmitted transitively.

[0169] While the technical concepts of the present disclosure have been described through various embodiments, the technical concepts of the present disclosure encompass various substitutions, modifications, and variations that can be made within the scope understandable to those of ordinary skill in the art to which the present disclosure pertains. Furthermore, it should be understood that such substitutions, modifications, and variations are encompassed within the scope of the appended claims.

Claims

1. In a method performed by an electronic device, A step of determining a prediction model to be analyzed among multiple prediction models that produce predicted sales for multiple items to analyze the cause of the error between the predicted sales and actual sales; A step of identifying an error cause analysis model corresponding to the above analysis target prediction model, wherein the error cause analysis model includes one or more decision rules identified based on the corresponding analysis target prediction model; and A method comprising a step of calculating error cause information corresponding to the analysis target prediction model based on an error cause analysis model corresponding to the analysis target prediction model.

2. In paragraph 1, The step of determining the prediction model to be analyzed above is: A step of calculating total error information regarding the predicted sales of the plurality of items based on the total predicted sales corresponding to the plurality of items and the total actual sales corresponding to the plurality of items calculated based on the plurality of prediction models; and A method comprising a step of determining the prediction model to be analyzed among the plurality of prediction models based on the total error information.

3. In paragraph 2, A method in which the above total error information is confirmed based on a weighted mean absolute percentage error (WMAPE) value between the total expected sales and total actual sales corresponding to the plurality of items.

4. In paragraph 1, The step of determining the prediction model to be analyzed above is: A step of calculating error contribution information corresponding to each of the above plurality of items; A step of identifying at least some items among the plurality of items for which the cause of the error is to be analyzed based on the error contribution information; and A method comprising the step of determining a prediction model corresponding to at least some of the identified items as the analysis target prediction model.

5. In paragraph 4, The step of calculating error contribution information corresponding to each of the above multiple items is: A step of calculating first total error information corresponding to all of the above multiple items; A step of calculating second total error information corresponding to one or more items excluding the target item that is the target of error contribution calculation among the above multiple items; and A method comprising a step of calculating error contribution information corresponding to the target item based on the first total error information and the second total error information.

6. In paragraph 4, A method wherein a prediction model corresponding to at least some of the identified items is determined based on at least some of item information, timing information, sales-related information, out-of-stock information, and promotion information corresponding to at least some of the identified items.

7. In paragraph 1, The step of determining the prediction model to be analyzed above is: A step of identifying an item at risk of stock shortage among the above plurality of items; and A method comprising the step of determining a prediction model corresponding to the above inventory shortage risk item as the analysis target prediction model.

8. In paragraph 7, The steps for identifying the above items at risk of stock shortage are: A step of calculating individual item error information corresponding to each of the plurality of items specified for each of the plurality of periods; A step of calculating the number of occurrences of underprediction errors among the plurality of periods based on the individual item error information; and A method comprising the step of identifying items at risk of stock shortage based on the number of occurrences of the above underprediction error and the total period corresponding to the entire plurality of periods.

9. In paragraph 7, The steps for identifying the above items at risk of stock shortage are: A step of calculating individual item error information corresponding to each of the plurality of items specified for each of the plurality of periods; A step of calculating the number of occurrences of underprediction errors exceeding a predetermined threshold error among individual item error information corresponding to the above multiple periods; and A method comprising the step of identifying a stock-out item based on the number of occurrences of underprediction errors exceeding the above threshold error.

10. In paragraph 7, The steps for identifying the above items at risk of stock shortage are: A step of calculating a stock shortage risk ranking of each of the plurality of items based on the number of occurrences of underprediction errors corresponding to each of the plurality of items; and A method comprising the step of identifying items at risk of stock shortage based on the above stock shortage risk ranking.

11. In paragraph 1, The above error cause analysis model is, A decision tree model including one or more rule nodes corresponding to each of the one or more decision rules, The step of calculating the above error cause information is: A step of determining a rule node corresponding to the expected sales amount calculated based on the analysis target prediction model among the one or more rule nodes based on the one or more decision rules; A step of identifying a result node corresponding to the analysis target prediction model based on the determined rule node and the calculated expected sales volume; and A method comprising a step of calculating error cause information corresponding to the above result node.

12. In paragraph 11, A level is assigned to each of the above one or more rule nodes, A method wherein the decision tree model comprises a rule node corresponding to a higher level and one or more rule nodes or one or more result nodes branched from the rule node corresponding to the higher level, wherein the branched one or more rule nodes or one or more result nodes correspond to a lower level adjacent to the higher level.

13. In paragraph 1, One or more decision rules corresponding to the above analysis target prediction model are: A method determined based on pattern information identified based on at least some of the predicted sales volume calculated based on the analysis target prediction model for a predetermined past period, the actual sales volume corresponding to the predetermined past period, and the error cause information calculated for the predetermined past period.

14. In electronic devices, One or more processors, comprising one or more memories storing instructions executed by the one or more processors; An electronic device, wherein when the instructions are executed by the one or more processors, the one or more processors are configured to execute a method according to any one of claims 1 to 13.

15. In a non-transitory computer-readable recording medium recording instructions that cause one or more processors to perform an operation when executed by one or more processors, A non-transitory computer-readable recording medium configured to cause the one or more processors to execute a method according to any one of claims 1 to 13.

Images