[3M1-GS-10] AI application

[3M1-GS-10-04] Factor analysis model for improving evaluation value by BERT and SHAP for review text data

〇Mamiko Watanabe¹, Koki Yamada¹, Ryotaro Shimizu¹, Satoshi Suzuki¹, Masayuki Goto¹ (1. Waseda University )

Keywords:Review text, BERT, Explainable AI, SHAP, Business Data Analysis

User ratings of accommodations on major booking sites are helpful information for travelers when making travel plans. They are also important indicators for accommodations because they can visualize the quality of their services. Furthermore, since review texts directly reflect users' impressions, it is possible to ascertain the factors that contribute to each users' satisfaction or dissatisfaction. Therefore, it would be effective if we could obtain an analytical method to derive and understand factors that influence the evaluation of accommodations. In this study, BERT converts all review text data in the dataset into high-dimensional vectors that reflect the context. Additionally, the obtained vectors are applied to several methods, such as clustering, regression, and explainable AI method. The proposed method enables to interpret the improvement points of each accommodation facility individually and contributes to the planning of marketing measures. We apply the proposed method to real-world service data and demonstrate its practical usage.