基于自监督学习的会话推荐算法研究  

Research on a Session Recommendation Algorithm Based on Self Supervised Learning

作　　者：陈景 叶维裕 CHEN Jing;YE Weiyu(Chongzuo Preschool Education College,Chongzuo Guangxi 532200;Guangxi Vocational Institute Of Technology,Chongzuo Guangxi 532200)

机构地区：[1]崇左幼儿师范高等专科学校,广西崇左532200 [2]广西理工职业技术学院,广西崇左532200

出　　处：《软件》2024年第8期10-13,共4页Software

基　　金：崇左幼儿师范高等专科学校2023年校级科研课题“智能技术赋能职业教育评价探索与研究——以计算机课程教学评价为例”(2023XB03)。

摘　　要：会话推荐旨在根据一段连续的交互序列给出合理的推荐。现有的会话推荐算法主要基于循环神经网络等,相较于传统方法取得了不错的效果。然而,这些方法只是考虑到了会话的序列信息,忽略了会话中的物品转移信息。为解决该问题,本文提出了基于图神经网络的会话推荐算法S-SRGNN,通过图神经网络建模物品之间的转移信息,捕获物品间的复杂关系。另外,在局部图和全局图上结合自监督学习,通过最大化不同视图会话表示的互信息加强会话建模。在三个公开数据集上的实验表明,S-SRGNN在会话推荐任务上取得了更好的效果。Session recommendation aims to provide reasonable recommendations based on a continuous sequence of interactions.The existing session recommendation algorithms are mainly based on recurrent neural networks and have achieved good results compared to traditional methods.However,these methods only consider the sequence information of the session and ignore the transfer information of items in the session.To solve this problem,this paper proposes a session recommendation algorithm S-SRGNN based on graph neural networks,which models the transfer information between items through graph neural networks and captures the complex relationships between items.In addition,combining self supervised learning on local and global graphs enhances session modeling by maximizing the interaction information between different view session representations.Experiments on three publicly available datasets have shown that S-SRGNN achieves better performance in session recommendation tasks.

关 键 词：计算机系统结构 神经网络 会话推荐 自监督学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]