Purchase Data Analysis Apparatus for Store Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing purchase data analysis methods struggle to accurately cluster stores with similar characteristics due to reliance on sales time and proceeds, which are influenced by merchandise assortment, stockout, and sales methods, leading to inaccurate sales predictions when stores with different characteristics are misclassified.
Innovation Solution
A purchase data analysis apparatus that acquires customer information on a customer-by-customer basis, generates customer representations based on action patterns, and then generates store representations to cluster stores effectively, reflecting customer habits and store characteristics, thereby reducing the influence of merchandise assortment and sales methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clustering is performed using commodity sales data, then the clustering result reflects differences in merchandise assortment and sales methods, but it becomes difficult to obtain proper clustering when merchandise assortment differs between stores and common commodities are few
Solution Approach 1:
The patent extracts and removes the influence of merchandise assortment differences from the clustering process. By using sales time and sales proceeds data that are independent of specific commodity types, the method extracts only the temporal and quantitative patterns of sales behavior, eliminating the confounding effect of different merchandise assortments across stores.
Solution Approach 2:
The patent changes the parameters used for clustering from commodity-specific parameters (commodity sales data) to temporal and quantitative parameters (sales time and sales proceeds). This parameter transformation allows clustering to be performed on a common basis across stores with different merchandise assortments, while still capturing meaningful store characteristics through the temporal patterns and sales volumes.
2Adaptability or versatility
If only sales time and sales proceeds are used for clustering, then the amount of information for clustering stores is limited, but the clustering can be performed without influence from merchandise assortment differences
Solution Approach 1:
The patent introduces temporal dimension (sales time patterns) as an additional dimension for clustering, complementing the quantitative dimension (sales proceeds). By analyzing the temporal patterns of sales activities alongside sales volumes, the method enriches the information available for clustering without requiring commodity-specific data, thus maintaining adaptability while reducing information loss.
3Reliability
If stores with different characteristics are misclassified into the same cluster, then sales prediction accuracy deteriorates, but achieving proper clustering requires more sophisticated methods
Solution Approach 1:
The patent segments the sales data into distinct temporal components (sales time patterns and sales proceeds) that can be analyzed separately and then combined for clustering. This segmentation allows the use of relatively simple clustering algorithms on each component while capturing different aspects of store behavior, achieving reliable clustering without requiring overly complex methods.
Data Source
AI summary
A purchase data analysis apparatus includes processing circuitry. The processing circuitry is configured to: acquire, on a customer-by-customer basis, customer information including an action time of a purchase-related action relating to purchase; generate, on a customer-by-customer basis, a customer representation representing an action pattern of a customer, based on the action time; generate, on a store-by-store basis, a store representation representing a representation of a customer coming to a store, based on the customer representation; and cluster stores by using the store representation.