Graph neural networks (GNNs) have recently shown impressive re- sults in recommender systems, achieving remarkable results across various sectors. However, their deployment in such systems is not without challenges. One significant issue is the initialization of random weights, where incorrect starting points can hinder the learning process, causing slow convergence or getting stuck in local minima. To address this, we introduce a novel two-phase method tailored to the unique demands of recommender systems. Initially, we refine the latent weights by aligning them with be- havioral pattern similarities, inspired by the collaborative filtering technique. Subsequently, we implement a new strategy for manag- ing the number of negative samples and devising an innovative loss function. These strategic modifications substantially boost GNN performance, aligning them more closely with the needs of recom- mender systems. This is evidenced by our method outperforming existing state-of-the-art methods on two different public datasets.
- Weighted loss function to prioritize informative training samples
- Adaptive margins to learn fine-grained user-item relationships
- Similarity-based embedding initialization for faster convergence
- Strategic edge removal techniques (bridge removal and noisy edge removal) to reduce computational complexity and improve embedding quality
- Comprehensive evaluation on benchmark datasets (Amazon-Book and Gowalla)
- Python 3.x
- PyTorch
- NumPy
- SciPy
- scikit-learn
- Clone the repository:
git clone https://github.com/OrKatz7/Multi-Aspect-Control-Graph-Recommendation-System
cd MultiAspectControlGraphRecommendationSystem- Prepare the dataset:
- Download the Amazon-Book and Gowalla datasets
- Preprocess the data and save it in the appropriate format, use preprocessing.py to initilaize the similiarty-based embedding matrix.
- Train:
run_amazon.shThis project is licensed under the MIT License.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent