ALDA4Rec: Adaptive Long-term Embedding for Recommendation
A new recommendation method, Adaptive Long-term Embedding with Denoising and Augmentation for Recommendation (ALDA4Rec), has been introduced. The model constructs an item-item graph, filters noise via community detection, and enriches user-item interactions. It uses Graph Convolutional Networks (GCNs) for short-term representations and averaging, GRUs, and attention mechanisms for long-term embeddings. An MLP-based adaptive weighting strategy dynamically optimizes long-term user preferences. Experiments on four real-world datasets show ALDA4Rec outperforms existing methods.
Key facts
- ALDA4Rec is a novel recommendation model.
- It constructs an item-item graph.
- Noise is filtered through community detection.
- User-item interactions are enriched.
- GCNs learn short-term representations.
- Long-term embeddings use averaging, GRUs, and attention.
- An MLP-based adaptive weighting strategy optimizes preferences.
- Tested on four real-world datasets.
Entities
—