Presentation Information
[2Yin-A-42]Classifying Restaurant Business Categories and Analyzing Regional Characteristics from Purchase Data Using LDA
〇Sota Maruyama1, Daiju Ibe2, Yuki Shigoku2, Ryo Taguchi1 (1. Nagoya Institute of Technology, 2. Infomart Corporation)
Keywords:
Latent Dirichlet Allocation (LDA),Topic Model,Regional Analysis,Purchase Data,Restaurant Business Categories
Restaurant category information is fundamental for understanding store characteristics in demand forecasting, marketing strategies, and regional analysis. While official classifications like the Japan Standard Industrial Classification exist, proprietary definitions by food service companies and reservation sites are widely used in practice, leading to a lack of standardization. Furthermore, categories are often self-reported, which may not reflect actual operations or changes over time. To support advanced decision-making, it is desirable to automatically assign category labels that reflect the actual state of business.In this study, we propose an unsupervised method for classifying restaurant categories from transaction data (purchase data) between restaurants and suppliers using Latent Dirichlet Allocation (LDA). We conducted an experiment using purchase data from 360 restaurants in Nagoya City from April 2022 to March 2025. The results demonstrated that our method yielded detailed classifications that better represented actual business characteristics and menu trends compared to conventional categories. Additionally, an analysis of the geographical distribution of the obtained LDA topics revealed distinct regional differences in their occurrence trends.