FlowNeg: GFlowNet-Guided Diverse Hard Negative Sampling for Knowledge Graph Embedding

Summary:ArXiv introduces FlowNeg, a novel method for negative sampling in Knowledge Graph Embedding (KGE) that utilizes a context-conditioned hierarchical Generative Flow Network (GFlowNet) to achieve reward-proportional sampling without normalizing the composite reward over the entity set. FlowNeg's terminal reward combines bounded model-based hardness with a training-only structural score to minimize collisions with held-out positives. Experiments demonstrate that FlowNeg significantly improves Mean R

---

Core Breakthroughs

• FlowNeg Method: FlowNeg is a novel negative sampling method based on GFlowNet, designed to address the inefficiency of negative sampling in Knowledge Graph Embedding (KGE).
• Reward Mechanism: FlowNeg employs a context-conditioned hierarchical generative flow network to achieve reward-proportional sampling without normalizing the composite reward over the entity set.
• Performance Improvement: Experimental results demonstrate that FlowNeg significantly improves Mean Reciprocal Rank (MRR) across multiple benchmarks while maintaining high diversity and low collision rates.

Technical Highlights

1. Hierarchical Generative Flow Network: FlowNeg leverages GFlowNet for negative sampling, effectively capturing complex data distributions.
2. Reward-Proportional Sampling: The method avoids normalizing the composite reward over the entity set, simplifying the sampling process.
3. Structural Scoring: Combines bounded model-based hardness with training-only structural scoring to minimize conflicts with held-out positives.

Industry Impact

• Knowledge Graph Applications: FlowNeg provides a more efficient negative sampling method for knowledge graph embedding, helping to enhance the performance of knowledge graph models.
• AI Research Community: This method offers new insights for the AI research community, particularly in handling large datasets and complex models.
• Developer Recommendations: Developers can apply FlowNeg to existing knowledge graph embedding models to improve accuracy and robustness.

Future Directions

FlowNeg demonstrates significant potential in the field of negative sampling. Future research could explore its applications in other domains, such as recommendation systems and image retrieval. Additionally, optimizing FlowNeg with other advanced technologies, such as reinforcement learning and deep learning, is a promising avenue for further exploration.

---

Source: ArXiv cs.LG (2026-08-24)

— END —

Tags: #FlowNeg #Knowledge Graph Embedding #GFlowNet #Negative Sampling #ArXiv