摘要
传统的协同过滤方法很大程度上以一种统一的方式进行建模,没有在交互意图的细粒度上考虑建模关系。为此提出一种融合交互意图的图神经网络协同过滤算法(INTNGCF)。首先,通过将边分解成多个潜在空间来识别潜在意图;其次,利用交互意图融合层确定这些潜在意图的重要性;最后,生成用户对项目的预测评分。在三个真实数据集上与七种基线模型进行对比实验,结果表明,提出的算法与其他先进的推荐模型相比具有一定的性能优势。
Traditional collaborative filtering methods largely model in a uniform way without considering the modeling relationships at the fine-grained level of interaction intent.This paper proposed a graph neural network collaborative filtering algorithm fused with interaction intention,named INTNGCF.Firstly,it identified the potential intention by decomposing edges into multiple potential spaces.Secondly,it used the interaction intention fusion layer to determine the importance of these potential intentions.Finally,it generated a predicted user rating of the project.Compared with seven baseline models on three real datasets,the results show that the proposed algorithm has certain performance advantages compared with other advanced recommendation models.
作者
张贤坤
秦锋斌
孙月
黄文杰
董梅
Zhang Xiankun;Qin Fengbin;Sun Yue;Huang Wenjie;Dong Mei(School of Artificial Intelligence,Tianjin University of Science&Technology,Tianjin 300457,China;Witeyesz Corp.,Ltd.,Shenzhen Guangdong 518029,China)
出处
《计算机应用研究》
CSCD
北大核心
2023年第2期488-492,共5页
Application Research of Computers
基金
天津市高等学校本科教学质量与教学改革研究计划项目(B201005706)
天津市自然科学基金资助项目(19JCYBJC15300)
天津市科技计划项目(21ZYQCSY00050)
天津市研究生科研创新项目(2021YJSS04)。
关键词
推荐系统
协同过滤
交互意图
图神经网络
recommendation system
collaborative filtering
interaction intent
graph neural network
