A Survey on Generative Knowledge Graph Construction
钊 洪
Artificial Intelligence and Robotics Research
Problems Identified (5)
KGC error propagation: Traditional discriminative knowledge graph construction pipelines suffer from error propagation.
KGC weak cross-domain generalization: Traditional knowledge graph construction pipelines have limited cross-domain generalization ability.
Generative KGC paradigm synthesis: Recent progress in generative knowledge graph construction requires systematic review across seq2seq and LLM-driven paradigms.
KGC error propagation: Traditional discriminative knowledge graph construction pipelines suffer from error propagation.
KGC weak cross-domain generalization: Traditional knowledge graph construction pipelines have limited cross-domain generalization ability.
Proposed Solutions (5)
Generative KGC survey: The paper surveys generative knowledge graph construction, covering classical seq2seq methods, LLM-based components, paradigm comparisons, and future directions.
Seq2Seq generative KGC: Generative KGC methods use end-to-end sequence-to-sequence modeling as an alternative to discriminative pipelines.
LLM-driven KGC: Large language models support full-process generative KGC across ontology construction, knowledge extraction, and knowledge fusion.
Generative KGC survey: The paper surveys generative knowledge graph construction, covering classical seq2seq methods, LLM-based components, paradigm comparisons, and future directions.
Seq2Seq generative KGC: Generative KGC methods use end-to-end sequence-to-sequence modeling as an alternative to discriminative pipelines.
Results (3)
Comprehensive generative KGC review:
Paradigm comparison:
Future directions for generative KGC:
Research Domain
Generative knowledge graph construction