New Mem-GF method slashes memory use for scalable collaborative filtering

Researchers have developed Mem-GF, a novel method for memory-efficient graph filtering in collaborative filtering (CF) that significantly reduces memory usage and improves runtime speed. Unlike previous methods that require storing the full item similarity graph, Mem-GF leverages Krylov subspaces to approximate polynomial graph filters without this explicit storage. Experiments show Mem-GF uses up to 5.74x less memory and offers a 4.38x speedup, while outperforming state-of-the-art GF and GCN-based methods in recommendation accuracy and scaling to large datasets. AI

IMPACT This new method could enable more scalable and efficient recommendation systems by overcoming memory limitations in graph-based approaches.

RANK_REASON Academic paper detailing a new method for collaborative filtering. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.IR (Information Retrieval) →

• arXiv
• collaborative filtering
• Graph convolutional networks
• Hugging Face
• Mem-GF

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →