An Autoencoder Framework With Attention Mechanism for Cross-Domain Recommendation

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommender systems face challenges like data sparsity and the cold-start problem.
• Existing cross-domain recommendation methods often rely on matrix factorization, limiting feature extraction.

Purpose of the Study:

• To propose a novel autoencoder framework with an attention mechanism (AAM) for enhanced cross-domain recommendation.
• To improve information transfer and fusion across different domains for accurate rating prediction.

Main Methods:

• Utilized autoencoder, multilayer perceptron, and self-attention for feature extraction and latent factor learning.
• Introduced AAM++ incorporating multihead self-attention for multi-aspect affinity learning between domains.

Main Results:

• The proposed AAM framework demonstrated superior performance over state-of-the-art methods in cross-domain recommendation.
• AAM++ showed improved results compared to AAM, particularly on sparse and large-scale datasets.

Conclusions:

• The AAM framework effectively transfers and fuses information across domains, enhancing recommendation accuracy.
• AAM++ offers a more robust solution for cross-domain recommendation, especially in challenging data scenarios.