DocSage: An Information Structuring Agent for Multi-Doc Multi-Entity Question Answering

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• DocSage significantly outperforms state-of-the-art long-context LLMs and RAG systems on MDMEQA benchmarks, achieving more than 27% accuracy improvements.result 90%
  Evidencepartial

  Evaluations on two MDMEQA benchmarks demonstrate that DocSage significantly outperforms state-of-the-art long-context LLMs and RAG systems, achieving more than 27% accuracy improvements respectively.

  Implicationpartial

  Directly stated in abstract with clear numeric improvement claim

  Verificationpartial

  partial

• Standard RAG's vector similarity-based coarse-grained retrieval often omits critical facts in multi-document multi-entity question answering.limitation 85%
  Evidencepartial

  standard RAG's vector similarity-based coarse-grained retrieval often omits critical facts

  Implicationpartial

  Directly stated as a limitation of existing approaches in the abstract

  Verificationpartial

  partial

• Graph-based RAG fails to efficiently integrate fragmented complex relationship networks in multi-document scenarios.limitation 85%
  Evidencepartial

  graph-based RAG fails to efficiently integrate fragmented complex relationship networks

  Implicationpartial

  Directly stated as a limitation of existing approaches in the abstract

  Verificationpartial

  partial

• DocSage uses a schema discovery module that dynamically infers query-specific minimal joinable schemas to capture essential entities and relationships.method 90%
  Evidencepartial

  A schema discovery module dynamically infers query-specific minimal joinable schemas to capture essential entities and relationships

  Implicationpartial

  Directly stated as a core component of the method in the abstract

  Verificationpartial

  partial

• DocSage transforms unstructured text into semantically coherent relational tables with error-aware correction mechanisms to reduce extraction errors.method 90%
  Evidencepartial

  An extraction module transforms unstructured text into semantically coherent relational tables, enhanced by error-aware correction mechanisms to reduce extraction errors

  Implicationpartial

  Directly stated as a core component of the method in the abstract

  Verificationpartial

  partial

• DocSage's reasoning module performs multi-hop relational reasoning over structured tables, leveraging schema awareness to efficiently align cross-document entities.method 90%
  Evidencepartial

  A reasoning module performs multi-hop relational reasoning over structured tables, leveraging schema awareness to efficiently align cross-document entities and aggregate evidence

  Implicationpartial

  Directly stated as a core component of the method in the abstract

  Verificationpartial

  partial

• DocSage offers precise fact localization via SQL-powered indexing as a key advantage over existing approaches.technical 85%
  Evidencepartial

  precise fact localization via SQL-powered indexing

  Implicationpartial

  Directly stated as an advantage of the proposed method in the abstract

  Verificationpartial

  partial

• DocSage mitigates LLM attention diffusion via structured representation, which is a key advantage of its agentic design.technical 85%
  Evidencepartial

  mitigated LLM attention diffusion via structured representation

  Implicationpartial

  Directly stated as an advantage of the proposed method in the abstract

  Verificationpartial

  partial

Startup potential card