메타패스 기반 이종 그래프 모델을 활용한 패션 커뮤니티 사용자의 패션 착장 선호도 연구

초록

Given the subjective nature of fashion, it is essential to consider user preferences when recommending fashion items. In this study, we introduce a graph-based model for look attribute preference modeling-specifically, a meta-path based heterogeneous graph deep learning model. This model utilizes outfit posts from social media to better understand user preferences. We gathered data from a representative online fashion community, Lookbook.nu, which includes 456,329 outfit images from 2,497 users, along with 328 extracted fashion attributes. A heterogeneous graph was created, with nodes representing users, item images, and fashion attributes.Using meta-path-based graph learning, we updated the multi-modal node features to reflect the relationships among the different node types. We then evaluated the effectiveness of our model in recommending fashion items based on user preferences. Our findings highlight the potential for analyzing personal preferences through social media posts to enhance recommender systems across various domains, including fashion.

키워드