Extracting User Segments Based on Specific Lifestyles Using Household Account Book Data

Abstract

Analyzing users' lifestyles based on household account book data is important for providing services tailored to users. However, the target lifestyle for service provision changes depending on the service provider's requirements. Therefore, there is a challenge where an approach that requires creating a model in advance, such as supervised learning, cannot adapt quickly to changes in requirements. In this study, we propose a method to extract users corresponding to any specified lifestyle without pre-modeling using a Large Language Model (LLM). However, this method has the problem that the search keywords generated by the LLM are not always suitable for the notational characteristics of household account book data. To improve this issue, the proposed method extends the search keywords using two embedding models, language and behavioral sequence. We analyzed the result of the proposed method through experiments using actual household account book data.