[2L1-GS-10t] AI application

9:15 AM - 9:30 AM JST(12:15 AM - 12:30 AM UTC)

[2L1-GS-10t-02] Analysis of ECRTM-Based Iterative Topic Refinement Using Large Language Models on Product Descriptions

〇Kei Takaba¹, Jiayi Wang¹, Tianxiang Yang², Tengfei Shao¹, Masayuki Goto¹ (1. Waseda University, 2. Keio University)

Keywords:

Topic model,ECRTM,Large Language Model,Topic Refinement,Product description

For external evaluators who aim to position competing products in the market, it is important to organize product groups using perspectives different from existing category systems. Grouping products based on semantic aspects such as functional characteristics and usage scenarios is effective for capturing users' fine-grained needs. However, as the number of products in e-commerce markets continues to grow rapidly, manually constructing such semantic-based categorizations has become increasingly difficult. Therefore, methods that can automatically extract new classification axes based on semantic aspects are required. To address this issue, applying topic models to product descriptions, which directly contain semantic information, is a promising approach. This study proposes a method that iteratively integrates the Embedding Clustering Regularization Topic Model with topic refinement using large language models. The effectiveness of the proposed method is demonstrated through experiments on real data, showing its ability to automatically extract highly interpretable classification axes.