.png){kind=icon}
返回
Multimodal heterogeneous graph convolutional network for image recommendation
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-023-01136-4.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
To improve the efficiency of the connection between people and information in specific scenarios, recent work has focused on mining user preferences from interactions. However, with the emergence of multimodal information in recent years, user choice in the image recommendation domain is influenced by multiple factors, such as image style, tags, and user social relationships, etc. Therefore, to explore user preferences under different modalities, we capture potential user preferences in a multimodal collaborative manner. In this work, a multimodal heterogeneous graph convolutional network model for image recommendation is proposed, which explores the differences in the representation of user preferences under different modalities. For different modalities, deep propagation networks are employed to construct higher-order connectivity coding between user heterogeneous interactions and image, tag, and user preference information. In addition, a dual-channel attention strategy with the idea of partitioning is employed to optimize the potential preferences of users. The experiments are conducted on public real-world datasets, the results clearly demonstrate the collaborative ability of multimodal information and heterogeneous interaction relations in exploring user preferences.
Keyword:
Image recommendation
Multimodal fusion
User personalization preferences
Dual-channel attention mechanism
期刊
IF:
3.1
论文数:
2.8K
被引数:
2.7K
机构
引用论文
Seismic anisotropy indicates organized melt beneath the Mid-Atlantic Ridge aids seafloor spreading.png)
Geology
OAAI
GeologyIF0
OAAI分享收藏
Hashtag our stories: Hashtag recommendation for micro-videos via harnessing multiple modalities
PREAI
分享收藏分享收藏A unified structure learning framework for graph attention networks图注意网络的统一结构学习框架NEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏 没有更多内容
没有更多内容