Researchers have developed FlowNeg, a novel method for generating diverse and informative negative samples in knowledge graph embedding (KGE) models. This approach utilizes a context-conditioned hierarchical generative flow network to amortize reward-proportional sampling, selecting entity types and then specific entities based on a combined reward of model-based hardness and a structural score that avoids collision with held-out positives. Experiments across various architectures and benchmarks show FlowNeg consistently outperforms existing methods like EMU and IF-NS in mean MRR, demonstrating its effectiveness in improving KGE model learning. AI
IMPACT Enhances knowledge graph embedding models by improving the quality and diversity of negative sampling, potentially leading to more accurate and robust KGE systems.
RANK_REASON The item describes a new method presented in a research paper for knowledge graph embedding. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- FB15K-237
- FlowNeg
- GFlowNets for AI-driven scientific discovery
- IF-NS
- knowledge graph embedding
- Rotate
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
