一种基于邻域与位置时间感知的会话推荐方法  

A Session Recommendation Method Based on Neighborhood and Position Time Awareness

作　　者：谭小辉 祁正华 张海桃 何菲菲 TAN Xiaohui;QI Zhenghua;ZHANG Haitao;HE Feifei(School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学计算机学院、软件学院、网络空间安全学院,江苏南京210023

出　　处：《软件导刊》2023年第11期91-97,共7页Software Guide

基　　金：国家自然科学基金项目(61972209)。

摘　　要：针对现有基于邻域的会话推荐方法忽略了整个会话的一般偏好、当前偏好以及匿名用户误点行为等问题,提出一种基于邻域与位置时间感知的会话推荐方法。该推荐方法考虑了3个因素,分别是项目在当前会话中的持续时间及位置、会话之间的时间相关性和邻居会话中候选项目的持续时间及位置。在Globo.com和Adressa两个公共新闻数据集上将该推荐方法与基准方法进行对比实验,结果表明,与表现最好的基准方法相比,该推荐方法在HR@20和NDCG@20指标上均有所提升。特别是在Globo.com数据集上,HR@20和NDCG@20指标分别提升了4.3%与7.0%,表明该方法能有效提升推荐效果。Aiming at the problem that the existing neighborhood-based session recommendation methods ignore the general preference,current preference,and anonymous user misclick behavior of the whole session.A session recommendation method based on neighborhood and position time awareness is proposed.The proposed recommendation method considers three factors,namely the duration and location of items in the current session,temporal correlation between sessions,and the lasting time and position of candidate items in neighboring sessions.Experimenting with the recommendation method with the benchmark method on two public news datasets,Globo.com and Adressa,the experimental results show that the recommendation method improves on both HR@20 and NDCG@20 metrics compared to the best-performing benchmark method.Especially on the Globo.com dataset,the HR@20 and NDCG@20 indicators were improved by 4.3%and 7.0%,respectively,indicating that the proposed method can effectively improve the recommendation effect.

关 键 词：会话推荐 位置感知推荐 时间相关性 误点行为 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]