Global-Aware External Attention Deep Model for Sequential Recommendation
File version
Author(s)
Wang, C
Griffith University Author(s)
Primary Supervisor
Other Supervisors
Editor(s)
Kashima, Hisashi
Ide, Tsuyoshi
Peng, Wen-Chih
Date
Size
File type(s)
Location
Osaka, Japan
License
Abstract
The sequential recommender plays a major role in contemporary recommendation systems, which shows the strong ability to model sequential patterns among the dataset. The classic sequential recommenders utilize the convolutional neural network, recurrent neural network, and self-attention mechanism to model the user’s preferences of items. However, these existing sequential recommendation models face the “Filter Bubble” issue by putting too much attention on each user’s own historical sequence, and they also ignore the feature-level item-item relationship. To address the existing challenges, we propose a novel global-aware external attention deep model (EDM) to learn both the global and local user preferences. The proposed EDM mainly contains a multi-embedding layer, an external attention layer, a feature-wise feed-forward network, and the candidate matching layer. Specifically, the external attention layer uses two external memory units shared across the entire input set to model the global interests of users. Then, by applying the feed-forward network to each feature dimension, the feature-wise feed-forward network is capable to learn the feature-level dependencies and properly model the local user preferences. In the experiments, three benchmark datasets are used with various validation metrics to show that our proposed EDM outperforms the state-of-the-art methods.
Journal Title
Conference Title
Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part III
Book Title
Edition
Volume
13937
Issue
Thesis Type
Degree Program
School
Publisher link
Patent number
Funder(s)
Grant identifier(s)
Rights Statement
Rights Statement
Item Access Status
Note
Access the data
Related item(s)
Subject
Data mining and knowledge discovery
Information and computing sciences
Persistent link to this record
Citation
Wang, T; Wang, C, Global-Aware External Attention Deep Model for Sequential Recommendation,Advances in Knowledge Discovery and Data Mining: 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2023, Osaka, Japan, May 25–28, 2023, Proceedings, Part III, 2023, 13937, pp. 335-347