Enhancing Recommendation Systems through SVD-based collaborative filtering and community detection.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Recommendation systems face challenges including data sparsity, scalability, and lack of interpretability.
• Classical Collaborative Filtering (CF) using Singular Value Decomposition (SVD) is scalable but struggles with extreme sparsity.
• Graph Neural Networks (GNNs) offer accuracy but lack explanatory power.

Purpose of the Study:

• To develop a hybrid recommendation framework addressing the trade-off between sparsity, interpretability, and scalability.
• To improve the performance of SVD-based collaborative filtering in sparse data scenarios.
• To create recommendation engines with enhanced transparency and reduced computational cost.

Main Methods:

• A novel hybrid framework sequentially combining Louvain community identification with SVD-based collaborative filtering.
• Utilizing modularity-based clustering to regularize user space partitioning for denser sub-matrices.
• Applying SVD on homogeneous sub-matrices derived from community clustering.

Main Results:

• The hybrid model achieved a Root-Mean-Square Error (RMSE) of 0.9966 and Mean Absolute Error (MAE) of 0.7968 on the Netflix Prize dataset.
• Demonstrated competitive predictive performance with significantly higher interpretability and lower computational cost compared to deep learning baselines.
• Successfully reduced computational noise and overhead, enabling community-level recommendation justifications.

Conclusions:

• The proposed hybrid framework effectively overcomes the sparsity-interpretability trade-off in recommendation systems.
• This approach enables scalable and transparent recommendation engines suitable for large-scale, sparse datasets.
• The method offers a practical solution for enhancing recommendation explainability and efficiency.