自注意力增强的动态个性化多行为推荐模型  

Namic personalized multi-behavior recommendation model with self-attention enhancement

作　　者：杨栩 曹琼[1] 黄贤英[1] 陈毓哲 YANG Xu;CAO Qiong;HUANG Xian-ying;CHEN Yu-zhe(School of Computer Science and Engineering,Chongqing University of Technology,Chongqing 400054,China)

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《计算机工程与设计》2025年第4期1134-1140,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62141201)。

摘　　要：为解决实例级建模中无法有效捕获用户个性化偏好和时序信息以及没有考虑用户对不同行为的差异性,提出一种融合时间元知识和注意力机制融合交互图的多行为推荐模型(MB-TMSCI)。在实例级多行为建模中纳入元学习范式,通过引入元知识个性化表示用户和物品嵌入;通过对交互时间编码考虑动态特征;利用多头注意力机制融合高阶图集且使用自注意力机制区分融合不同类型的高阶图集。在3个公开数据集上进行大量实验,验证了所提模型的推荐效果优于基准模型。To address the inability to effectively capture users’personalized preferences and the temporal information in instance-level modeling and the lack of consideration of users’variability of behaviors,a multi-behavior recommendation model,MB-TMSCI,fusing temporal meta-knowledge and attention mechanisms to address multi-behavior recommendation tasks effectively was proposed.Within the instance-level multi-behavior modeling,a meta-learning paradigm was used to introduce meta-know-ledge to personalize the representation of user and item embeddings.Dynamic features were considered by encoding the interaction timestamps.A multi-head attention mechanism was used to fuse high-order graph sets and the self-attention mechanisms was employed to distinguish and merge different high-order graph sets.Extensive experiments on three public datasets verify that the proposed model outperforms the benchmark in terms of recommendation.

关 键 词：多行为推荐 时间编码 元知识 高阶交互 注意力机制 图神经网络 显式建模 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]