Search

KR-20260063145-A - METHOD AND APPARATUS FOR SKU CLUSTERING BASED ON SEARCH KEYWORDS

KR20260063145AKR 20260063145 AKR20260063145 AKR 20260063145AKR-20260063145-A

Abstract

An SKU clustering method according to one embodiment of the present disclosure is performed by a computing device and comprises: a step of setting a number of clusters to classify a plurality of keywords; a step of inputting a plurality of keyword-related information and the number of clusters into a first model to map each of the plurality of keywords to one of the plurality of clusters; a step of identifying a category to which a first SKU belongs and a keyword corresponding to the category to which the first SKU belongs; a step of mapping a first SKU to one of the plurality of clusters based on the keyword corresponding to the category to which the first SKU belongs; and a step of predicting the demand for the first SKU based on the mapping result.

Inventors

  • 시몬슨 나단 시드니
  • 찬다르 델리프쿠마르

Assignees

  • 쿠팡 주식회사

Dates

Publication Date
20260507
Application Date
20241030

Claims (14)

  1. In a method performed by a computing device, A step of setting the number of multiple clusters to classify multiple keywords; A step of inputting multiple keyword-related information and the number of multiple clusters into a first model, and mapping each of the multiple keywords to one of the multiple clusters; A step of identifying a category to which a first SKU belongs, and a keyword corresponding to the category to which the first SKU belongs; A step of mapping the first SKU to any one of the plurality of clusters based on a keyword corresponding to the category to which the first SKU belongs; and A step comprising predicting the demand for the first SKU based on the mapping result, SKU clustering method.
  2. In Article 1, The above clusters are distinguished in relation to the season, The demand for the above-mentioned first SKU is demand during a specific period related to the season, and The above-mentioned first SKU is characterized by belonging to a fashion-related category, SKU clustering method.
  3. In Article 1, The above multiple keyword-related information includes multiple keywords and multiple keyword-related sales data, and The above multiple keywords are keywords that have an association with fashion exceeding a pre-set threshold, and The above multiple keywords have undergone a data smoothing process, and The above multiple keyword-related sales data is characterized by being time-series based data. SKU clustering method.
  4. In Article 1, The step of setting the number of clusters above is, A step of identifying the above plurality of keyword-related information; A step of performing K-means modeling based on the above plurality of keyword-related information; and It includes a step of deriving the optimal number of clusters identified through the above K-means modeling, and The above K-means modeling is a process of identifying the optimal number of clusters through n-cluster simulation, and The above optimal number of clusters is characterized by being dynamically adjusted, SKU clustering method.
  5. In Article 1 The step of mapping each of the above plurality of keywords to any one of the above plurality of clusters is A step of performing time series analysis on the above plurality of keyword-related information; A step of mapping each of the plurality of keywords to one of the plurality of clusters based on the above time series analysis results; A step of identifying the correlation coefficient between the above keyword and the cluster mapped to the above keyword; and A step comprising identifying keywords having a suitability greater than a preset threshold for each cluster based on correlation coefficients, SKU clustering method.
  6. In Article 5, The step of identifying keywords having a suitability greater than a preset threshold for each of the above clusters is: Step of extracting characteristics of each cluster; A step of extracting the characteristics of keywords mapped to each cluster; A step of performing exploratory data analysis (EDA) on the characteristics of each cluster and the characteristics of keywords mapped to each cluster; and It includes a step of identifying keywords having a suitability of greater than or equal to a preset threshold for each cluster, and The characteristics of each of the above clusters and the characteristics of the keywords mapped to each of the above clusters include trends, seasonality, or randomness identified through seasonal trend decomposition, and The above suitability is characterized by being determined through keyword frequency, correlation coefficient, or similarity. SKU clustering method.
  7. In Article 1, The demand for the above-mentioned first SKU is predicted by taking into account the attributes of the above-mentioned first SKU, and The attributes of the first SKU include the sales volume, inventory status, product category, precipitation-related status, or holiday-related status of the first SKU, and The attributes of the first SKU are characterized by being identified based on time series analysis. SKU clustering method.
  8. One or more processors; and It includes memory that stores computer programs executed by one or more processors, and When the above computer program is executed, the above one or more processors: An operation to set the number of multiple clusters to classify multiple keywords, The operation of inputting multiple keyword-related information and the number of the multiple clusters into a first model, and mapping each of the multiple keywords to one of the multiple clusters. The operation of identifying the category to which the first SKU belongs, and the keyword corresponding to the category to which the first SKU belongs, An operation of mapping a first SKU to any one of the plurality of clusters based on a keyword corresponding to the category to which the first SKU belongs, and Performing the operation of predicting the demand for the first SKU based on the mapping result, SKU clustering device.
  9. In Article 8, The above clusters are distinguished in relation to the season, The demand for the above-mentioned first SKU is demand during a specific period related to the season, and The above-mentioned first SKU is characterized by belonging to a fashion-related category, SKU clustering device.
  10. In Article 8, The above multiple keyword-related information includes multiple keywords and multiple keyword-related sales data, and The above multiple keywords are keywords that have an association with fashion exceeding a pre-set threshold, and The above multiple keywords have undergone a data smoothing process, and The above multiple keyword-related sales data is characterized by being time-series based data. SKU clustering device.
  11. In Article 8, The operation of setting the number of clusters above is, The operation of identifying the above multiple keyword-related information, The operation of performing K-means modeling based on the above plurality of keyword-related information, and It includes an operation to derive the optimal number of clusters identified through the above K-means modeling, and The above K-means modeling is a process of identifying the optimal number of clusters through n-cluster simulation, and The above optimal number of clusters is characterized by being dynamically adjusted, SKU clustering device.
  12. In Article 8, The operation of mapping each of the above plurality of keywords to any one of the above plurality of clusters is, The operation of performing time series analysis on the above multiple keyword-related information, The operation of mapping each of the plurality of keywords to one of the plurality of clusters based on the above time series analysis results, An operation to identify the correlation coefficient between the above keyword and the cluster mapped to the above keyword, and Includes an operation to identify keywords having a suitability greater than a preset threshold for each cluster based on correlation coefficients, SKU clustering device.
  13. In Article 12, The operation of identifying keywords having a suitability of greater than or equal to a preset threshold in each of the above clusters is, Operation to extract characteristics of each cluster, Operation to extract characteristics of keywords mapped to each cluster, An operation to perform exploratory data analysis (EDA) on the characteristics of each cluster and the characteristics of keywords mapped to each cluster, and It includes an operation to identify keywords having a suitability of greater than a preset threshold for each cluster, and The characteristics of each of the above clusters and the characteristics of the keywords mapped to each of the above clusters include trends, seasonality, or randomness identified through seasonal trend decomposition, and The above suitability is characterized by being determined through keyword frequency, correlation coefficient, or similarity. SKU clustering device.
  14. In Article 8, The demand for the above-mentioned first SKU is predicted by taking into account the attributes of the above-mentioned first SKU, and The attributes of the first SKU include the sales volume, inventory status, product category, precipitation-related status, or holiday-related status of the first SKU, and The attributes of the first SKU are characterized by being identified based on time series analysis. SKU clustering device.

Description

Method and apparatus for SKU clustering based on search keywords The present disclosure relates to a method and apparatus for clustering stock keeping units (SKUs) based on search keywords. Specifically, it may relate to SKU clustering technology including the process of setting the number of clusters, mapping keywords to clusters, and mapping specific SKUs to clusters. In modern business, SKU (Stock Keeping Unit) demand forecasting is essential for efficient inventory management and sales growth. Inaccurate SKU demand forecasting can lead to issues such as excess or shortage inventory, which can significantly impact a company's profitability. This is particularly critical in the fashion industry, where sales fluctuate significantly due to seasonal variations and specific events. The fashion category includes many products heavily influenced by seasonality and trends, leading to a tendency for demand to concentrate on specific items each season. For instance, demand for cold-weather gear such as coats, puffer jackets, and scarves surges during the winter season, while conversely, demand for lightweight items like short-sleeved t-shirts and sandals rises in the summer. Furthermore, specific events such as holidays, vacation seasons, and various promotional periods influence customer purchasing patterns, causing shifts in product demand. These factors highlight the need for companies to implement sophisticated demand forecasting systems that go beyond simple inventory management. Existing technologies have primarily predicted SKU demand based on inventory levels and customer data regarding additional stock requests. However, this approach has fundamental limitations. For example, even if demand for a product is predicted to increase, if inventory is insufficient or out of stock (OOS), sales opportunities for that product are lost, leading to revenue loss for the company. In particular, for products with strong seasonality, if adequate inventory is not secured within a limited season, the company may not only fail to secure expected sales but also experience a significant decline in customer satisfaction. With the recent advancement of e-commerce platforms, analysis utilizing search data is becoming increasingly important for accurately understanding customer purchasing intent. By analyzing customer search patterns, the correlation between specific keywords and products, and search trends over specific periods, it is possible to more clearly identify what products customers are looking for and at what point they are considering a purchase. In particular, if companies can predict SKU demand by reflecting changes in demand driven by seasonal characteristics and events, they will be able to maximize sales through more efficient inventory management and personalized product recommendations. To address these issues, a new SKU demand forecasting method that reflects customer purchasing behavior and seasonal factors is required. FIG. 1 is a drawing illustrating an SKU clustering-related system according to one embodiment of the present disclosure. Figure 2 is a diagram illustrating the K-means Elbow Curve, which shows the process of determining the optimal number of clusters through the K-means algorithm. Figure 3 is a diagram illustrating the results of analyzing search patterns based on seasonality and non-seasonality by clustering various keywords. Figure 4 is a diagram showing a pie chart visualizing the proportion of each cluster in the total keywords as a result of clustering various keywords. Figure 5 is a diagram illustrating the process of clustering multiple keywords and analyzing the accuracy and reliability of the clustering by evaluating the results using correlation coefficients. Figure 6 illustrates the process of measuring the goodness of fit between keywords and clusters using the Pearson Correlation Coefficient and evaluating the reliability of clustering through this. Figure 7 is a diagram illustrating the process of identifying suitable keywords corresponding to the category to which a specific SKU belongs. Figure 8 is a diagram illustrating the results of analyzing the search/customer index of non-seasonal clusters, that is, keywords that are searched steadily throughout the year without being significantly affected by seasonal factors. Figure 9 is a diagram illustrating the results of analyzing the search activities of keywords belonging to seasonal clusters by period. FIG. 10 is a flowchart illustrating an SKU clustering method according to one embodiment of the present disclosure. FIG. 11 is a flowchart illustrating the detailed steps of step S1010 of FIG. 10. FIG. 12 is a flowchart illustrating the detailed steps of step S1020 of FIG. 10. FIG. 13 is a flowchart illustrating the detailed steps of step S1240 of FIG. 12. FIG. 14 is a block diagram showing the hardware configuration of a computing device for inventory health management according to one embodiment of the present disclosure. Prefer