Get Started
Home
Topics
Search
Library
Research questionHow much complementary knowledge can link-prediction models recover together in incomplete knowledge graphs?Different link-prediction models and training runs can produce substantially different answers for the same knowledge-graph query. Combining them may recover information missed by individual models, but some queries may remain unresolved collectively.
AI
Evaluation & Benchmarks
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.Oracle, will I ever learn? A study of prediction convergence and complementarity across link prediction modelsThe study examines multiple link-prediction architectures and benchmarks. Its oracle selects the best prediction from each considered model for every query, estimating an upper bound on combination performance; the results show substantial complementarity, rapid saturation as models are added, and a persistent set of unsolved queries.research paper · Sep 2, 2026
Related questions
How can knowledge graph completion predict links for unseen entities while adapting path relevance to each query’s structural context?How can uncertain knowledge graphs predict missing confidence scores without losing global structure or training stability?How can knowledge graph embeddings model valid inference patterns without over-generalizing from sparse evidence?How can entity-alignment models transfer to unseen heterogeneous knowledge graphs without retraining?