摘要
Abstract
User preference representation is one of the core tasks of recommender systems,and its accuracy and comprehensive-ness directly affects the quality of recommendation results and user experience.User behavior-driven preference representation has attracted much attention because it directly reflects users'real interests.By analyzing users'historical behavior data,such as clicks,browsing,purchases,and ratings,recommender systems can construct representations that reflect users'preference.This paper aims to provide a comprehensive analysis and exposition of user behaviour-driven preference representation in recom-mender systems from three aspects:information sources of preference representation,single-vector preference representation and multi-vector preference representation.Specifically,starting from the information sources of preference representation,this paper respectively explores the representation based on interaction item features and the representation based on comment text,and analyzes the roles and common methods of these two types of information in preference representation.Then,from the per-spectives of single-vector representation and multi-vector representation,it analyzes the advantages and disadvantages of differ-ent representation methods and their application scenarios.Finally,it discusses the development trends of user behaviour-based preference representation in recommendation models,aiming to provide ideas and directions for subsequent research.关键词
推荐系统/用户偏好表征/用户行为/表征方法/偏好建模Key words
recommender systems/user preference representation/user behavior/representation methods/preference modeling分类
信息技术与安全科学
