融合时间序列特征的群组推荐模型  

A group recommendation model integrating time series features

作　　者：朱欣娟 熊依伦 ZHU Xinjuan;XIONG Yilun(School of Computer Science/The Shaanxi Key Laboratory of Clothing Intelligence,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学计算机科学学院/陕西省服装设计智能化重点实验室,陕西西安710048

出　　处：《西安工程大学学报》2024年第1期105-112,120,共9页Journal of Xi’an Polytechnic University

基　　金：国家重点研发计划项目(2019YFC1521400)。

摘　　要：针对传统的群组推荐预定义策略过于单一,忽视用户与项目之间的交互性,无法捕捉时间推移所造成的用户偏好迁移等问题,提出一种融合时间序列和注意力机制的群组推荐模型TAGR(time-attitation group rememdation)。首先通过层次聚类划分出高相似度群组,其次引入时间序列模型来捕捉用户偏好迁移过程,获取每个时刻用户行为的兴趣偏好,并聚合各时刻兴趣偏好作为用户偏好。最后结合注意力机制,获得用户权重进行偏好融合来表示群组偏好,最终作为推荐模型的输入。通过在Goodbook与MovieLens数据集上与NCF、AGREE等模型进行对比,TAGR在归一化折扣累计增益和命中率2个指标上都得到了显著提高。Traditional group recommendation has such problems as ineffective predefined strategy,neglect of the interaction between users and projects,and failure to capture the migration of user preferences caused by the passage of time.In response to the above problems,a group recommendation model TAGR(time-attentive group recommendation)that integrates time series and attention mechanisms was proposed.Firstly,high similarity groups were divided through hierarchical clustering.Secondly,a time series model was introduced to capture the process of user preference transfer,obtain the interest preferences of user behavior at each moment,and aggregate the interest preferences at each moment as user preferences.Finally,with attention mechanism,user weights were obtained for preference fusion to represent group preferences,serving as the input of the recommendation model.By comparing with NCF,AGREE and other models on the Goodbook and MovieLens datasets,the proposed model TAGR has been significantly improved in both normalized discount cumulative gain and hit rate.

关 键 词：群组推荐 时间序列 层次聚类 神经网络 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]