Method for classifying products, and device therefor

Abstract

The present disclosure relates to a method for classifying products, and a device therefor. The product classification method according to one embodiment of the present disclosure may comprise the steps of: acquiring time-series sales volume data for each of a plurality of products; extracting features associated with sales patterns of each of the plurality of products by using the sales volume data; and classifying the plurality of products into at least one segment on the basis of the features associated with the sales patterns, wherein the features associated with the sales patterns include a seasonality feature, and the step of classifying into at least one segment can include the step of classifying, according to a first index calculated from the seasonality feature, the plurality of products into a first segment and a second segment that differs from the first segment.

Description

Method and device for classifying goods

[0001] The present disclosure relates to a method and a device for classifying products, and more specifically, to a method and a device for classifying products into a plurality of segments based on characteristics of the sales patterns of the products and providing a demand prediction model specialized for each segment.

[0002] As Internet use becomes more widespread, the e-commerce market is expanding, and the transaction volume and items of e-commerce are continuously growing.

[0003] Meanwhile, e-commerce businesses analyze historical inventory trends and patterns, and based on this analysis, strive to maintain appropriate inventory levels in fulfillment centers. For example, they use demand forecasting models, generated through learning from time-series data on product sales volume, to predict product demand and manage inventory accordingly.

[0004] However, even if products belong to the same category, sales patterns, seasonal characteristics, trends, etc. may be completely different, so there is a problem that the results may be distorted if a single demand forecasting model is applied to all products.

[0005] Accordingly, the inventor of the present invention seeks to provide a method for forming clusters for products with similar sales patterns and improving the accuracy of demand forecasting by using a demand forecasting model specialized for each cluster.

[0006] A technical problem to be solved through embodiments of the present disclosure is to provide a method and system for forming a cluster of products with similar sales patterns by classifying products into a plurality of segments according to the characteristics of the sales patterns.

[0007] A technical problem to be solved through other embodiments of the present disclosure is to improve the accuracy of demand forecasting by using a demand forecasting model specialized for each segment to support efficient inventory management.

[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] According to one embodiment of the present disclosure for solving the above-described technical problem, a product classification method includes the steps of obtaining sales volume data in a time-series format for each of a plurality of products, the step of extracting a feature associated with a sales pattern of each of the plurality of products using the sales volume data, and the step of classifying the plurality of products into at least one segment based on the feature associated with the sales pattern, wherein the feature associated with the sales pattern includes a seasonal feature, and the step of classifying the plurality of products into at least one segment may include the step of classifying the plurality of products into a first segment and a second segment different from the first segment according to a first indicator calculated from the seasonal feature.

[0010] In one embodiment, the step of obtaining the sales data may include a step of collecting sales data within a certain period from the current time and a step of filtering the data based on whether the first sale point of the product is within the recent period or whether there is no sales history for the recent period.

[0011] In one embodiment, the step of extracting features associated with the sales pattern may include the step of preprocessing the sales amount data using a smoothing value generated from a Gaussian filter.

[0012] In one embodiment, the preprocessing step may include replacing outliers due to promotions associated with the plurality of products with corresponding smoothing values.

[0013] In one embodiment, the preprocessing step may include detecting outliers in the sales data and replacing the detected outliers with corresponding smoothing values.

[0014] In one embodiment, the step of detecting the outlier may include the step of generating residual data using data smoothed with a Gaussian filter.

[0015] In one embodiment, the step of extracting features associated with the sales pattern may include the step of normalizing the numerical units of the sales amount data.

[0016] In one embodiment, the step of extracting features associated with the sales pattern may include the step of transforming the sales data into a frequency domain using a Fourier transform.

[0017] In one embodiment, the first indicator derived from the seasonal feature is an amplitude of a pattern representing the seasonal feature, and the step of classifying into at least one segment may include a step of determining whether the amplitude is greater than or equal to a first threshold.

[0018] In one embodiment, the method may further include a step of reclassifying products classified into the first segment based on average sales volume per segment.

[0019] In one embodiment, the reclassifying step may include: estimating a seasonal section and an off-season section from sales data of products classified into the first segment; comparing the average sales of the seasonal section with the average sales of the off-season section; and classifying products classified into the first segment into a third segment and a fourth segment different from the third segment based on the comparison result.

[0020] According to other embodiments of the present disclosure for solving the above-described technical problem, a demand prediction method includes the steps of: obtaining segment information of a target product using sales data for a certain period of time for the target product; selecting a demand prediction model based on the segment information; and predicting demand for the target product using the demand prediction model, wherein the step of selecting a demand prediction model includes the steps of: selecting a first demand prediction model as the demand prediction model when the target product is classified into a first segment; and selecting a second demand prediction model different from the first demand prediction model as the demand prediction model when the target product is classified into a second segment, wherein the first demand prediction model may be learned using sales data of a plurality of products classified into the first segment, and the second demand prediction model may be learned using sales data of a plurality of products classified into the second segment.

[0021] According to other embodiments of the present disclosure for solving the above-described technical problem, a computing device includes a processor and a memory storing instructions, wherein the instructions, when executed by the processor, cause the processor to: obtain sales data in a time-series format for each of a plurality of products; extract a feature associated with a sales pattern of each of the plurality of products using the sales data; and classify the plurality of products into at least one segment based on the feature associated with the sales pattern, wherein the feature associated with the sales pattern includes a seasonal feature, and the classifying into at least one segment may include a step of classifying the plurality of products into a first segment and a second segment different from the first segment according to a first indicator calculated from the seasonal feature.

[0022] According to other embodiments of the present disclosure for solving the above-described technical problem, a computing device includes a processor and a memory storing instructions, wherein the instructions, when executed by the processor, cause the processor to: obtain segment information of a target product using sales data for a certain period of time for the target product; select a demand prediction model based on the segment information; and predict demand for the target product using the demand prediction model, wherein the operation of selecting a demand prediction model includes: selecting a first demand prediction model as the demand prediction model when the target product is classified into a first segment; and selecting a second demand prediction model, different from the first demand prediction model, as the demand prediction model when the target product is classified into a second segment, wherein the first demand prediction model may be learned using sales data of a plurality of products classified into the first segment, and the second demand prediction model may be learned using sales data of a plurality of products classified into the second segment.

[0023] FIG. 1 is a diagram exemplarily showing the configuration of an inventory management system according to one embodiment of the present disclosure.

[0024] FIG. 2 is a flowchart of a product classification method according to another embodiment of the present disclosure.

[0025] FIG. 3 is a diagram exemplarily showing sales data for products filtered according to one embodiment of the present disclosure.

[0026] Figure 4 is a detailed flowchart for explaining step S200 illustrated in Figure 2.

[0027] Figures 5 to 7 are exemplary drawings for explaining some of the operations illustrated in Figure 4.

[0028] Figure 8 is a detailed flowchart for explaining step S300 illustrated in Figure 2.

[0029] FIG. 9 and FIG. 10 are exemplary drawings for explaining some of the operations illustrated in FIG. 8.

[0030] Figure 11 is a detailed flowchart for explaining step S400 illustrated in Figure 2.

[0031] Figure 12 is a drawing for explaining some of the processes shown in Figure 11.

[0032] FIG. 13 is a flowchart of a demand prediction method according to another embodiment of the present disclosure.

[0033] FIG. 14 is a block diagram showing the hardware configuration of a computing device for product classification according to an embodiment of the present disclosure.

[0034] Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the attached drawings. The advantages and features of the present disclosure, and methods for achieving them, will become clear with reference to the embodiments described in detail below together with the attached drawings. However, the technical idea of ​​the present disclosure is not limited to the following embodiments and may be implemented in various different forms. The following embodiments are provided only to complete the technical idea of ​​the present disclosure and to fully inform those skilled in the art of the present disclosure of the scope of the present disclosure, and the technical idea of ​​the present disclosure is defined only by the scope of the claims.

[0035] In describing various embodiments of the present disclosure, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present disclosure, the detailed description will be omitted.

[0036] Unless otherwise defined, the terms (including technical and scientific terms) used in the following examples may be used with meanings commonly understood by those of ordinary skill in the art to which this disclosure pertains; however, this may vary depending on the intentions of engineers working in the relevant field, precedents, the emergence of new technologies, etc. The terminology used in this disclosure is for the purpose of describing the embodiments and is not intended to limit the scope of this disclosure.

[0037] In the following examples, singular expressions include plural concepts unless the context clearly specifies that they are singular. Furthermore, plural expressions include singular concepts unless the context clearly specifies that they are plural.

[0038] In addition, terms such as first, second, A, B, (a), (b), etc. used in the following embodiments are only used to distinguish certain components from other components, and the nature, order, or sequence of the components are not limited by the terms.

[0039] Hereinafter, various embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0040] FIG. 1 is a diagram exemplarily illustrating an inventory management system according to one embodiment of the present disclosure. As illustrated in FIG. 1, the inventory management system according to one embodiment of the present disclosure may be configured to include a product classification system (10), and in some embodiments, may further include at least some of a database (20) and a user terminal (30).

[0041] The database (20) may be implemented as one or more computer systems that collect, accumulate, and / or generate identifier information for each management unit of multiple products, sales data in time series format, and promotional record information / promotion date information associated with the products. In some embodiments, the database (20) may be implemented as one or more computer systems that collect, accumulate, and / or generate information regarding products classified into different segments, information regarding segments, and information regarding demand forecasting models corresponding to the segments.

[0042] At least some of the information described above may be collected, processed, and / or managed from various activities at various fulfillment centers, and may be collected, processed, and / or managed individually by each fulfillment center. While the various pieces of information within the database (20) have been described above, the described content does not limit the scope of the information, and it is understood that the information may be varied within the scope of achieving the objectives of the present disclosure.

[0043] Meanwhile, in FIG. 1, the database (20) is depicted as a system external to the product classification system (10), but at least some of the functions and configurations of the database (20) may be implemented within the product classification system (10). Accordingly, at least some of the information described above may be managed in an internal database (not shown) within the product classification system (10).

[0044] The product classifier (100) can classify multiple products into at least one segment based on their sales patterns. For example, the product classifier (100) can obtain sales volume data in time-series format for each of the multiple products, extract features associated with the sales patterns of each of the multiple products using the sales volume data, and classify the multiple products into at least one segment based on the features associated with the sales patterns. In this case, the features associated with the sales patterns include seasonal features, and the product classifier (100) can classify the multiple products into a first segment and a second segment different from the first segment based on a first indicator derived from the seasonal features.

[0045] The demand predictor (200) can select a corresponding demand prediction model using segment information of the target product, and can predict the demand for the target product using the selected demand prediction model. For example, if the target product is classified into the first segment, the demand predictor (200) can select a first demand prediction model as the demand prediction model, and if the target product is classified into the second segment, the demand predictor (200) can select a second demand prediction model different from the first demand prediction model as the demand prediction model. In this case, the first demand prediction model may be learned using sales data of a plurality of products classified into the first segment, and the second demand prediction model may be learned using sales data of a plurality of products classified into the second segment.

[0046] Users can receive demand forecast information predicted by the demand forecaster (200) using a user terminal (30) and utilize the information to establish a plan for maintaining an appropriate inventory level or to establish a marketing plan. Meanwhile, the user terminal (30) may be implemented as a variety of computing devices, such as a smartphone, laptop, or desktop, and may be implemented as any device.

[0047] Meanwhile, the components illustrated in Fig. 1 can communicate via a network. Here, the network can be implemented as any type of wired / wireless network, such as a local area network (LAN), a wide area network (WAN), a mobile radio communication network, or Wibro (Wireless Broadband Internet).

[0048] The configuration and operation of the product classification system (10) according to the present embodiment have been described with reference to FIG. 1. The embodiments described above may be understood in more detail with reference to other embodiments described below. Furthermore, the technical concepts understood through the above-described embodiments may be applied to other embodiments described below, even if not specifically stated otherwise.

[0049] Next, a product classification method according to another embodiment of the present disclosure will be described with reference to FIGS. 2 through 12. The product classification method according to this embodiment may be performed by one or more computing devices. Furthermore, the product classification method according to this embodiment may be performed in part by a first computing device, and the remaining operations by a second computing device.

[0050] For example, some operations of the product classification method according to the present embodiment may be performed by an on-premise physical server, while the remaining operations may be performed by a cloud server. Hereinafter, if the performer of each operation is omitted, it will be understood that the performer is the computing device.

[0051] First, a product classification method according to the present embodiment will be briefly described with reference to FIG. 2. FIG. 2 is a flowchart of a product classification method according to another embodiment of the present disclosure.

[0052] According to this embodiment, sales volume data for each of a plurality of products is collected (S100). Specifically, hourly sales volume data for each of the plurality of products may be collected within a certain period of time from the present time. For example, sales volume data for a period extending up to three years prior to the time at which sales volume data for the plurality of products is collected (queried) may be obtained. In one embodiment, the plurality of products may include identifiers of different management units. In other words, the plurality of products may represent items distinguished by identifiers of management units. The identifiers of the management units may be, for example, a Stock Keeping Unit (SKU) or a Product ID.

[0053] According to the present embodiment, in step S100, some of the sales volume data for multiple products may be filtered according to preset criteria. Specifically, sales volume data for products with a short sales period (Virtually New SKU) or no recent sales history (Virtually Retired SKU) may be excluded in subsequent steps. This is because it is difficult to obtain meaningful information from sales volume data for products with a short sales period or no recent sales history. Therefore, by excluding such products prior to the sales volume data analysis step, the reliability and efficiency of data analysis are improved.

[0054] FIG. 3 is a diagram exemplarily showing sales data for products filtered according to one embodiment of the present disclosure. FIG. 3a shows the sales volume of Product A over a three-year period, and FIG. 3b shows the sales volume of Product B over a three-year period.

[0055] Referring to Figure 3a, we can see that Product A's first sale occurred within the past year. In this case, due to the insufficient amount of data related to Product A's sales, it is difficult to extract sales pattern characteristics from this data. Consequently, sales data for Product A are excluded in subsequent steps. In other words, if a product's first sale occurred within a certain period of time, sales data for that product can be filtered.

[0056] Referring to Figure 3b, we can see that Product B has no sales history within the past year. In this case, since Product B is no longer sold to new customers, sales data for Product B are excluded in subsequent steps. In other words, if a product has no sales history for a certain period of time, sales data for that product can be filtered.

[0057] Referring back to FIG. 2, features associated with the sales patterns of each of the multiple products are extracted using the sales data collected in step (S100) (S200), and based on the features associated with the extracted sales patterns, the multiple products can be classified into at least one segment (S300). Specifically, seasonality features of the sales patterns of each of the multiple products are extracted, and based on a first indicator derived from the seasonality features, the multiple products can be classified into a first segment and a second segment. This will be described in detail with reference to the drawings.

[0058] Figure 4 is a detailed flowchart of a method for extracting features associated with the sales pattern illustrated in Figure 2. Here, the features associated with the sales pattern may include seasonal characteristics of the sales pattern, and the seasonal characteristics may refer to the volatility of sales volume over a certain period.

[0059] Referring to Figure 4, preprocessing can be performed on the collected sales data (S211). This involves comparing the time-series sales data with surrounding data values ​​to detect outliers (sales spikes or outliers due to product promotions) that deviate from the general pattern and replacing them with smoothing values. To this end, a Gaussian filter can be applied to each sales data point to generate a smoothing value for each sales data point. Specifically, a one-dimensional Gaussian filter can be generated, and the smoothing value can be derived by calculating the weighted average of the surrounding data for each data point. This method allows for the generation of natural replacement values ​​similar to the pattern of the original sales data, since the smoothing value is derived by considering the preceding and following sales data for a specific point.

[0060] Next, we describe the process of detecting outliers (e.g., sales spikes) caused by product promotions in collected sales data and replacing them with smoothing values. When there are promotions, such as special events or discounts, associated with a product, sales can increase dramatically. Sales data resulting from such events can interfere with extracting characteristics of the product's overall sales pattern. Therefore, smoothing values ​​are necessary. To achieve this, promotional information associated with the product can be obtained. This promotional information can include the date of the promotion. Points corresponding to the promotional dates are detected in the product sales data, and the values ​​at those points are replaced with the smoothing values.

[0061] Next, the process of detecting outliers in the collected sales data and replacing them with smoothing values ​​is explained with reference to the drawing.

[0062] First, residual data can be generated using data smoothed with a Gaussian filter. Specifically, referring to FIG. 5, residual data (503) can be generated by subtracting data (502) smoothed with a Gaussian filter from time-series sales data (501). At this time, since the Gaussian filter smoothes the data using a kernel with a normal distribution form that places a high weight on the center point, the residual data using this also has a form close to a normal distribution. In other words, the distribution (histogram) of the residual data is formed in a symmetrical form centered around 0 (see FIG. 6).

[0063] Next, outliers can be detected using the Z-score method, and the data values ​​corresponding to the outliers in the original sales data can be replaced with the smoothing value described above. As previously explained, since the distribution of the residual data is symmetrical around 0, outlier detection using the Z-score method can be performed effectively. Since the method of detecting outliers using the Z-score method is a well-known technique, a detailed description is omitted.

[0064] In this way, in this embodiment, by performing preprocessing on sales volume data prior to extracting seasonal characteristics of sales patterns, a more accurate sales pattern can be identified for each product.

[0065] Referring back to FIG. 4, normalization can be performed on the preprocessed sales data (S220). The numerical units (or scales) of the time-series sales data for each product may differ significantly. Here, in order to focus on the patterns and trends of the time-series sales data rather than the numerical values ​​themselves, the data are normalized to resolve the differences in the numerical units (or scales). For example, Min-Max Normalization can be used to convert the time-series sales data into a common numerical unit between 0 and 1. However, the present embodiment is not limited thereto, and other known methods can be used to convert the numerical units of the data into numerical units within a certain range, such as p to q.

[0066] Next, a process for removing trend features from the sales data can be performed (S230). For example, this can be accomplished using linear regression or a method that converts annual sales growth into a slope to obtain a linear slope.

[0067] Next, the time-series sales data is converted into the frequency domain, from which seasonal characteristics can be extracted (S240). Here, seasonal characteristics can refer to a pattern of sales increases and decreases that repeat at regular intervals. Specifically, the time-domain sales data can be converted into frequency-domain data using a Fourier transform. By considering DC characteristics, the extracted seasonal characteristics can be intended to center around 0, thereby generating data (seasonality) that necessarily passes through 0 (see Figure 7).

[0068] Referring again to FIG. 2, multiple products can be classified into a first segment and a second segment based on a first indicator derived from a feature associated with a sales pattern, i.e., a seasonal feature for the sales pattern (S300).

[0069] Figure 8 is a detailed flowchart for explaining step S300 illustrated in Figure 2.

[0070] Referring to Figure 8, the amplitude of seasonal characteristics can be measured as a reference indicator (first indicator) for segment classification (S310). Here, the amplitude of seasonal characteristics refers to the range from the minimum to the maximum of the periodic increase / decrease pattern derived from the seasonal characteristics (see Figure 9).

[0071] Next, it can be determined whether the measured amplitude is greater than or equal to a first threshold (S320). If the amplitude is greater than or equal to the first threshold (YES), the product can be classified into the first segment (S330). If the amplitude is less than the first threshold (NO), the product can be classified into the second segment (S340).

[0072] Figure 10 is a diagram exemplifying data for products classified into first and second segments. Figure 10a shows data for product A, and Figure 10b shows data for product B. Furthermore, the seasonality characteristic may refer to the periodic increase / decrease pattern (Seasonality) depicted in Figure 10.

[0073] Referring to Figure 10a, if the amplitude derived from the seasonality feature extracted from the sales data of Product A is greater than or equal to the first threshold, Product A is a product with high sales volatility according to a certain cycle and can be classified into the first segment (Seasonal SKU).

[0074] Referring to Figure 10b, if the amplitude derived from the seasonality feature extracted from the sales data of Product B is less than the first threshold, Product B is a product that is less affected by a specific time or season and maintains a constant volume of sales throughout the year, and can be classified into the second segment (Non-Seasonal SKU).

[0075] Meanwhile, referring to FIG. 2, the product classification method according to the present embodiment may further include a step (S400) of reclassifying products classified into the first segment into the third and fourth segments. Sales data for products belonging to the first segment may be divided into a period in which demand significantly increases (seasonal period) and a period in which demand does not increase (off-season period). Depending on the product, some products may show a certain level of sales even in the off-season period and some may not. Products that show a certain level of sales even in the off-season period and products that have almost no sales in the off-season period need to be handled differently from an inventory management perspective. Therefore, in the present embodiment, products classified into the first segment may be reclassified into the third and fourth segments by comparing the average sales volumes of the seasonal and off-season periods. This will be described in detail below with reference to the drawings.

[0076] Figure 11 is a detailed flowchart for explaining step S400 illustrated in Figure 2.

[0077] To compare average sales volumes for seasonal and off-season periods in sales data for products classified as the first segment, the seasonal and off-season periods must first be distinguished in the sales data. To achieve this, pattern points can be extracted from seasonal feature data extracted from the sales data (S410).

[0078] Figure 12 is a drawing for explaining some of the processes shown in Figure 11.

[0079] As illustrated in Figure 12, points with meaningful information in seasonal feature data can be extracted as pattern points. For example, a first point (Local Max Point) indicating a local maximum point, a second point (Local Min Point) indicating a local minimum point, a third point (Inflection Point) indicating a point where the second derivative value of the extracted seasonal feature becomes 0, and a fourth point (Zero Cross Point) indicating a point where the extracted seasonal feature intersects 0 can be extracted.

[0080] Next, the seasonal and non-seasonal sections can be estimated using pattern points (S420). For example, the distance between the third point (Inflection Point) located to the left and right of the first point (Local Max Point) and the distance between the fourth point (Zero Cross Point) located to the left and right of the first point (Local Max Point) are measured, and the section with the larger value can be estimated as the seasonal section. At this time, if the second point (Local Min Point) is detected first during the process of searching the points located to the left and right of the first point (Local Max Point), the distance of the section can be measured by taking the second point (Local Min Point) into consideration.

[0081] For another example, the interval between the first point (Local Max Point) and the fourth point (Zero Cross Point) located to the left and right can be estimated as a season interval. Here, as in the previous example, if the second point (Local Min Point) is detected first during the process of searching for points located to the left and right of the first point (Local Max Point), the interval distance can be measured by taking the second point (Local Min Point) into account.

[0082] Next, the average sales volume for each season and off-season period can be measured (S430). Specifically, the average sales volume for the season period and the average sales volume for the off-season period within the past year can be measured.

[0083] Next, based on the results of comparing the measured average sales volume, the product can be classified into a third segment and a fourth segment (S440). Specifically, the ratio of the average sales volume in the seasonal period to the average sales volume in the off-season period is calculated, and if the ratio between the calculated average sales volumes is above a preset threshold, the product is classified into the third segment, and if it is below the threshold, the product is classified into the fourth segment. Here, the third segment represents a type in which most sales occur only in the seasonal period, and the fourth segment represents a type in which a certain level of sales occurs in sections other than the seasonal period (off-season period).

[0084] While we've described how products are classified based on seasonality, other characteristics extracted from sales patterns can also be used to classify products. For example, products can be classified based on the continuity of their sales patterns. Specifically, the average demand interval is calculated for each sales data point, and based on this result, products with discontinuous and intermittent sales patterns can be classified from those with continuous sales patterns.

[0085] Meanwhile, as previously explained, sales patterns, including sales patterns, seasonality, and trends, vary across products. Therefore, predicting demand using a single demand forecasting model based on sales data for all products can result in distorted forecast results. Therefore, in this disclosure, demand for a target product can be predicted using a demand forecasting model selected based on segment information for the target product. This will be described below.

[0086] FIG. 13 is a flowchart of a demand prediction method according to another embodiment of the present disclosure.

[0087] Referring to Figure 13, first, segment information into which the target product is classified can be obtained (S500). Here, the segment information of the target product refers to information regarding the segment into which the target product is classified according to the product classification method described in the preceding examples. For more information, please refer to the descriptions in the preceding examples.

[0088] Next, a demand prediction model may be selected based on segment information (S600). Here, the selected demand prediction model is a demand prediction model corresponding to the segment information, and may be learned using sales data of products in the segment into which the target product is classified. Specifically, if the target product is classified into the first segment, a demand prediction model (first demand prediction model) learned using sales data of multiple products classified into the first segment may be selected as a model for predicting demand for the target product, and if the target product is classified into the second segment, a demand prediction model (second demand prediction model) learned using sales data of multiple products classified into the second segment may be selected as a model for predicting demand for the target product.

[0089] Next, demand for the target product can be predicted using the selected demand prediction model (S700). Because the selected demand prediction model is specialized for the sales pattern characteristics of the segment in which the target product is classified, it can more accurately predict demand for the target product.

[0090] FIG. 14 is a block diagram showing the hardware configuration of a computing device for product classification according to an embodiment of the present disclosure.

[0091] Referring to FIG. 14, a computing device (500) may include one or more processors (510), a bus (530), a communication interface (540), a memory (520) for loading a computer program executed by the processor (510), and a storage (550) for storing a computer program (560). However, only components related to the embodiment of the present disclosure are illustrated in FIG. 10. Therefore, a person skilled in the art to which the present disclosure pertains will appreciate that other general components may be included in addition to the components illustrated in FIG. 14. That is, the computing device (500) may further include various components in addition to the components illustrated in FIG. 14. In addition, in some cases, the computing device (500) may be configured in a form in which some of the components illustrated in FIG. 9 are omitted. Hereinafter, each component of the computing device (500) will be described.

[0092] The processor (510) can control the overall operation of each component of the computing device (500). The processor (510) can be configured to include at least one of a Central Processing Unit (CPU), a Micro Processor Unit (MPU), a Micro Controller Unit (MCU), a Graphics Processing Unit (GPU), or any other type of processor well known in the art of the present disclosure. In addition, the processor (510) can perform operations for at least one application or program for executing operations / methods according to embodiments of the present disclosure. The computing device (500) can include one or more processors.

[0093] Next, the memory (520) can store various data, commands, and / or information. The memory (520) can load a computer program (560) from the storage (550) to execute operations / methods according to embodiments of the present disclosure. The memory (520) may be implemented as a volatile memory such as RAM, but the present disclosure is not limited thereto.

[0094] Next, the bus (530) can provide communication functions between components of the computing device (500). The bus (530) can be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

[0095] Next, the communication interface (540) can support wired and wireless Internet communication of the computing device (500). Furthermore, the communication interface (540) can also support various communication methods other than Internet communication. To this end, the communication interface (540) can be configured to include a communication module well known in the technical field of the present disclosure.

[0096] Next, the storage (550) can non-temporarily store one or more computer programs (560). The storage (550) can be configured to include non-volatile memory such as Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which the present disclosure pertains.

[0097] Next, the computer program (560) may include one or more instructions that, when loaded into the memory (520), cause the processor (510) to perform operations / methods according to various embodiments of the present disclosure. That is, the processor (510) may perform operations / methods according to various embodiments of the present disclosure by executing the loaded one or more instructions.

[0098] Various embodiments of the present disclosure and effects according to the embodiments have been described with reference to FIGS. 1 through 14. The effects according to the technical concept 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 below.

[0099] Furthermore, even though the above embodiments have described multiple components as being combined or operating in combination, the technical concept of the present disclosure is not necessarily limited to these embodiments. That is, within the scope of the technical concept of the present disclosure, all of the components may be selectively combined and operated one or more times.

[0100] The technical concepts of the present disclosure described so far can be implemented as computer-readable code on a computer-readable recording medium. A computer program recorded on the computer-readable recording medium can be transmitted to another computing device via a network such as the Internet, installed on the device, and used therein.

[0101] Although the operations are depicted in a specific order in the drawings, it should not be understood that the operations must be performed in the specific order depicted, or in a sequential order, or that all depicted operations must be performed to achieve the desired result. In certain circumstances, multitasking and parallel processing may be advantageous. Although various embodiments of the present disclosure have been described above with reference to the attached drawings, those skilled in the art to which the present disclosure pertains will understand that the technical concepts of the present disclosure can be implemented in other specific forms without changing the technical concepts or essential characteristics thereof. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. The scope of protection of the present disclosure should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included in the scope of the technical ideas defined by the present disclosure.

[0102] [Explanation of symbols]

[0103] 10: Product Classification System

[0104] 100: Product Classifier

[0105] 200: Demand Forecaster

[0106] 20: Database

[0107] 30: User terminal

[0108] S100, S200, S210, S220, S230, S240, S300, S310, S320, S330, S340, S400, S410, S420, S430, S440, S500, S600, S700: Steps

[0109] 501, 502, 503: Data

[0110] 500: Computing Device

[0111] 510: Processor

[0112] 520: Memory

[0113] 530: Bus

[0114] 540: Communication Interface

[0115] 550: Storage

[0116] 560: Computer program

Claims

1. In a method performed on a computing device, A step of obtaining sales volume data in time series format for each of multiple products; A step of extracting features associated with the sales pattern of each of the plurality of products using the above sales volume data; and Comprising a step of classifying the plurality of products into at least one segment based on characteristics associated with the above sales pattern, The characteristics associated with the above sales pattern include seasonal characteristics, The step of classifying into at least one segment includes a step of classifying the plurality of products into a first segment and a second segment different from the first segment according to a first indicator derived from the seasonality characteristic. How to classify products.

2. In paragraph 1, The step of acquiring the above sales data is a step of collecting sales data within a certain period from the present time; and Including a step of filtering based on whether the first sale point of the product is within a recent period of time or whether there is no sales history within a recent period of time. How to classify products.

3. In paragraph 1, The step of extracting features associated with the above sales pattern is: A step of preprocessing the above sales data using a smoothing value generated from a Gaussian filter, How to classify products.

4. In paragraph 3, The above preprocessing step is, Comprising a step of replacing outliers due to promotions associated with the above multiple products with corresponding smoothing values. How to classify products.

5. In paragraph 3, The above preprocessing step is, A step of detecting outliers in the above sales data, and comprising a step of replacing the detected outliers with corresponding smoothing values; How to classify products.

6. In paragraph 5, The step of detecting the above outliers is: A step of generating residual data using data smoothed with a Gaussian filter, How to classify products.

7. In paragraph 1, The step of extracting features associated with the above sales pattern is: Including a step of normalizing the numerical unit of the above sales data, How to classify products.

8. In paragraph 1, The step of extracting features associated with the above sales pattern is: Comprising a step of converting the sales data into a frequency domain using a Fourier transform, How to classify products.

9. In paragraph 1, The first indicator derived from the above seasonal characteristics is the amplitude of the pattern representing the above seasonal characteristics, The step of classifying into at least one segment is: Including a step of determining whether the above amplitude is greater than or equal to a first threshold value, How to classify products.

10. In paragraph 1, Further comprising a step of reclassifying the products classified into the first segment based on the average sales volume by segment. How to classify products.

11. In paragraph 10, The above reclassification step is, A step of estimating seasonal and non-seasonal sections from sales data of products classified into the above first segment; A step of comparing the average sales volume of the above-mentioned season period with the average sales volume of the above-mentioned off-season period; and A step of classifying the products classified into the first segment based on the comparison results into a third segment and a fourth segment different from the third segment, How to classify products.

12. In a method performed on a computing device, A step of obtaining segment information of a target product by using sales data for a certain period of time for the target product; A step of selecting a demand prediction model based on the above segment information; and A step of predicting demand for the target product using the demand prediction model is included, The step of selecting the demand prediction model includes a step of selecting a first demand prediction model as the demand prediction model when the target product is classified into the first segment, and a step of selecting a second demand prediction model different from the first demand prediction model as the demand prediction model when the target product is classified into the second segment. The first demand prediction model is learned using sales data of a plurality of products classified into the first segment, and the second demand prediction model is learned using sales data of a plurality of products classified into the second segment. Demand forecasting methods.

13. Processor; and Contains memory that stores instructions, The above instructions, when executed by the processor, cause the processor to: An action to obtain sales data in time series format for each of multiple products; An operation of extracting features associated with the sales pattern of each of the plurality of products using the above sales data; and Including an operation of classifying the plurality of products into at least one segment based on characteristics associated with the above sales pattern, The characteristics associated with the above sales pattern include seasonal characteristics, The operation of classifying into at least one segment includes a step of classifying the plurality of products into a first segment and a second segment different from the first segment according to a first indicator derived from the seasonality characteristic. Computing device.

14. Processor; and Contains memory that stores instructions, The above instructions, when executed by the processor, cause the processor to: An action of obtaining segment information of a target product by using sales data for a certain period of time for the target product; An operation of selecting a demand prediction model based on the above segment information; and Includes an action of predicting demand for the target product using the above demand prediction model, The operation of selecting the demand prediction model includes an operation of selecting a first demand prediction model as the demand prediction model when the target product is classified into the first segment, and selecting a second demand prediction model different from the first demand prediction model as the demand prediction model when the target product is classified into the second segment. The first demand prediction model is learned using sales data of a plurality of products classified into the first segment, and the second demand prediction model is learned using sales data of a plurality of products classified into the second segment. Computing device.

Images