摘要
针对基于匿名用户的会话推荐忽略了不同会话之间可能存在的协作信息,以及未考虑所预测的目标项与历史行为的相关性问题,提出一种基于局部-邻域图信息与注意力机制的会话推荐模型(SR-LNG-AM)。从当前会话和邻域会话构建的图结构中分别学习两种类型的项目转换信息,将其融合得到项目嵌入。使用软注意力机制生成全局嵌入,使用目标注意力机制针对不同的目标项自适应生成不同的目标嵌入。结合局部嵌入,进行预测。在两个真实数据集上与多个基线方法进行实验对比,实验指标均有提高,验证了该方法的有效性。
Session recommendation based on anonymous users ignores the possible cooperation information between different sessions and does not consider the correlation between the predicted target items and historical behavior,a session recommendation based on local-neighborhood graph information and attention mechanism(SR-LNG-AM)was proposed.The two types of project transformation information were learned from the graph structure constructed using the current session and neighborhood session,and they were fused to obtain the project embedding.The soft attention mechanism was used to generate the global embedding,and the target aware attention mechanism was used to adaptively generate different target embeddings for different target items.The local embedding was fused for prediction.Experiments were carried out on two real datasets.Compared with multiple baseline methods on two real datasets,the experimental indicators are all improved,which verifies the effectiveness of this method.
作者
党伟超
吴非凡
高改梅
刘春霞
白尚旺
DANG Wei-chao;WU Fei-fan;GAO Gai-mei;LIU Chun-xia;BAI Shang-wang(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)
出处
《计算机工程与设计》
北大核心
2024年第3期925-931,共7页
Computer Engineering and Design
基金
太原科技大学研究生教育创新基金项目(SY2022063)
太原科技大学博士科研启动基金项目(20202063)
山西省自然科学基金项目(201901D111266、201901D111252)。
关键词
会话推荐
注意力机制
图信息
邻域会话
协作信息
目标注意力
目标嵌入
session-based recommendation
attention mechanism
graph information
neighborhood session
collaboration information
target attention
target embedding
