计算机工程与应用2025，Vol.61Issue(21)：15-29,15.DOI:10.3778/j.issn.1002-8331.2501-0190
面向推荐系统的用户兴趣建模综述
Survey of User Interest Modeling for Recommendation Systems
摘要
Abstract
Focusing on the task of user interest modeling,this paper summarizes and analyzes the methods of click inten-tion identification and interest construction,and discusses the existing challenges in this field.User interest modeling includes two cascade stages:click intention recognition and interest construction.According to whether or not differential attention is paid to the features involved in user click behavior,click intention recognition methods are classified into two categories:personalized and non-personalized.According to the different processing methods of user click intention sequence,the interest construction method is divided into two categories:aggregation and generation,which provides clear research ideas for this field.Experiments are carried out on ml-20m and Amazon_all_beauty datasets,and Recall,Precision,MRR and NDCG are used as evaluation indicators to verify the advantages and disadvantages of various inter-est construction methods.User interest modeling can construct interest representation according to behavior sequence and its context information,help the model learn the implicit relationship between user behaviors,and then realize personal-ized recommendation service.However,this task still faces some challenges,such as personalized click intention recogni-tion method does not fully explore the potential correlation between unary click intentions,interest construction phase needs a deep understanding of interest diversity in order to capture different granularity of user interests and so on.关键词
推荐系统/兴趣建模/用户点击意图Key words
recommendation systems/interest modeling/user click intention分类
计算机与自动化引用本文复制引用
吕学强,王夏雨,马登豪..面向推荐系统的用户兴趣建模综述[J].计算机工程与应用,2025,61(21):15-29,15.基金项目
国家自然科学基金(62171043,62202061) （62171043,62202061）
北京市自然科学基金(4232025). （4232025）
