Abstract
Current LLM-based research agents have made strides in agent orchestration but largely overlook scientific knowledge orchestration. Existing works often reduce papers to abstracts, surface mentions, and flat \texttt{cites} edges, omitting key entities, claims, evidence, mechanisms, and method lineages essential for scientific reasoning. To this end, we introduce Agents-K1, an end-to-end knowledge orchestration pipeline that converts raw documents into agent-native scientific knowledge graphs.
Key Components
Agents-K1 integrates three components under a unifying theoretical foundation:
- Multimodal Parser: A five-module schema that captures entities, multimodal evidence, citations, and typed inter-entity relations across the full paper rather than abstracts alone.
- Information-extraction Backbone: A 4B information-extraction backbone trained with GRPO under a rule-based reward.
- Graphanything CLI: A tri-source agent interface that unifies web search, multimodal graph retrieval, and cross-document traversal.
Applications and Experiments
On top of this, we processed 2.46 million scientific papers across six subjects to produce Scholar-KG, releasing a one-million-paper subset, with the full Scholar-KG accessible via the SCP link below. The same pipeline can be extended to general-domain corpora and schema-conformant data synthesis. Extensive experiments demonstrate that Agents-K1 achieves superior performance in scientific information extraction, knowledge graph construction, and multi-hop scientific reasoning.
Blogger's Review: The introduction of Agents-K1 represents a significant advancement in the field of scientific knowledge orchestration, particularly in information extraction and knowledge graph construction, greatly enhancing the potential of agents in scientific research. Its design of a multimodal parser and extraction strategy provides valuable insights for future research.