{"slug": "flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph", "title": "FlowNeg: GFlowNet-Guided Diverse Hard Negative Sampling for Knowledge Graph Embedding", "summary": "Researchers introduced FlowNeg, a context-conditioned hierarchical generative flow network for negative sampling in knowledge graph embedding, which outperformed existing methods EMU and IF-NS in 24 of 25 benchmark cells, achieving a mean MRR improvement of +0.0172 and +0.0160 respectively. In a separate 15-seed control on FB15k-237/RotatE, FlowNeg achieved an MRR of 0.359±0.001 versus 0.346±0.002 for EMU, demonstrating higher diversity and gradient informativeness without treating structural similarity as an open-world truth oracle.", "body_md": "arXiv:2608.23849v1 Announce Type: new\nAbstract: Negative sampling determines whether a knowledge graph embedding (KGE) model learns from informative counterexamples or wastes updates on implausible corruptions. Uniform negatives are diverse but easy, whereas hard-negative miners concentrate on few entities and collide more with held-out positives. We introduce FlowNeg, a context-conditioned hierarchical generative flow network that amortizes reward-proportional sampling without normalizing a composite reward over the entity set: given a positive triple and corruption side, it selects a type, then an entity. Its terminal reward combines bounded model-based hardness with a training-only structural score for held-out-positive collision, over a relation-specific type-compatible support. We derive the reward, specialize standard trajectory balance, and bound multiplicatively how residual imbalance perturbs terminal and mode probability. Across a descriptive five-seed grid of five architectures and five benchmarks, FlowNeg has higher mean MRR than EMU and than IF-NS in 24 of 25 cells ($+0.0172$ and $+0.0160$ on average). A separate 15-seed FB15k-237/RotatE control fixing negative count, diagnostic budget, and compute gives FlowNeg $0.359\\pm0.001$ MRR against $0.346\\pm0.002$ for EMU, with near-uniform fixed-partition diversity, high gradient informativeness, and low collision. The evidence supports mode-covering negative generation without treating structural similarity as an open-world truth oracle.", "url": "https://wpnews.pro/news/flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph", "canonical_source": "https://arxiv.org/abs/2608.23849", "published_at": "2026-08-26 04:00:00+00:00", "updated_at": "2026-08-26 04:13:18.310698+00:00", "lang": "en", "topics": ["machine-learning", "artificial-intelligence"], "entities": ["FlowNeg", "EMU", "IF-NS", "FB15k-237", "RotatE"], "alternates": {"html": "https://wpnews.pro/news/flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph", "markdown": "https://wpnews.pro/news/flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph.md", "text": "https://wpnews.pro/news/flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph.txt", "jsonld": "https://wpnews.pro/news/flowneg-gflownet-guided-diverse-hard-negative-sampling-for-knowledge-graph.jsonld"}}