삼중항 손실 기반 사용자 분석을 통한 추천 시스템 개선

초록

A service provider must understand the unique characteristics of the users to recommend appropriate products. Factors such as personality, social class, and hobbies can be used to identify the user and improve recommendations. This information can be obtained through direct questioning or analyzing user-generated content such as blog posts. This study uses a large language model and transfer learning to extract specific information, such as taste and knowledge, from text. The author’s characteristics are then embedded using a recurrent neural network and a triplet loss metric learning methodology to distinguish between users. Qualitative insights into their similarities were then obtained from analyzing the label distribution in the embedding space.

키워드