Skip to content

Add ComoRAG (AAAI'26) to Hybrid GraphRAG and Multi-round Retriever#23

Open
EternityJune25 wants to merge 2 commits into
DEEP-PolyU:mainfrom
EternityJune25:main
Open

Add ComoRAG (AAAI'26) to Hybrid GraphRAG and Multi-round Retriever#23
EternityJune25 wants to merge 2 commits into
DEEP-PolyU:mainfrom
EternityJune25:main

Conversation

@EternityJune25
Copy link
Copy Markdown

Add ComoRAG (AAAI 2026) to two categories:

1. Hybrid GraphRAG

  • ComoRAG — Organizes knowledge into three complementary layers:
    • Veridical Layer: raw text + LLM-extracted knowledge triples with entity linking
    • Semantic Layer: hierarchical summaries via RAPTOR clustering
    • Episodic Layer: narrative flow summaries
      Combines textual and relational knowledge sources. Achieves consistent relative gains up to 11% on four long-narrative benchmarks (200K+ tokens).
      Paper

2. Multi-round Retriever

  • ComoRAG — Core innovation: a Metacognitive Control Loop
    Self-Probe → Tri-Retrieve → Mem-Encode → Mem-Fuse → Try-Answer
    Enables stateful, iterative retrieval and reasoning across multiple rounds with a dynamic memory workspace.
    Achieves consistent relative gains up to 11% on four long-narrative benchmarks (200K+ tokens).
    Paper

 Add [ComoRAG](https://ojs.aaai.org/index.php/AAAI/article/view/40644)(AAAI 2026) to two categories:
1. **Hybrid GraphRAG** — ComoRAG organizes knowledge into three complementary layers: Veridical Layer (raw text + LLM-extracted knowledge triples with entity linking), Semantic Layer (hierarchical summaries via RAPTOR clustering), and Episodic Layer (narrative flow summaries). This combines textual and relational knowledge sources.
2. **Multi-round Retriever** — The core innovation is a Metacognitive Control Loop (Self-Probe → Tri-Retrieve → Mem-Encode → Mem-Fuse → Try-Answer) enabling stateful, iterative retrieval and reasoning across multiple rounds with a dynamic memory workspace.
ComoRAG achieves consistent relative gains up to 11% on four long-narrative benchmarks (200K+ tokens).
**Paper:** https://doi.org/10.1609/aaai.v40i39.40644
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant