检索说明

  • 查询:GraphRAG evidence retrieval
  • 年份过滤:2025-
  • 单查询上限:20
  • 发现日期:2026-11-12

Semantic Scholar 候选元数据

  • 作者:Harshverdhan Soni, Parikshit Saikia, Badal Soni, Jitesh Choudhary
  • 年份:2026
  • Venue:2026 IEEE Guwahati Subsection Conference (GCON)
  • DOI:10.1109/GCON69192.2026.11649060
  • arXiv:未返回
  • S2 citationCount(当前快照):0
  • 来源线索:https://doi.org/10.1109/GCON69192.2026.11649060

摘要(未核验)

Large Language Models (LLMs) have revolutionized content generation but still face issues related to biases and factual foundation. Retrieval-Augmented Generation (RAG) solves the problem of grounding by conditioning generation on externally retrieved knowledge; however, standard pipelines use dense retrieval by mapping queries and documents into vector embeddings followed by flat ranking, resulting in redundant evidence and inefficient long-context processing. GraphRAG enhances retrieval through structured graph-based reasoning, yet does not explicitly account for token budget constraints. We propose GCP-GraphRAG, a token-aware graph-based retrieval framework that formulates evidence selection as a constrained optimization problem. Graph Context Pruning (GCP) selects compact, coherent subgraphs under fixed token budgets, enabling scalable multilingual and multimodal retrieval-augmented generation with improved context efficiency.

转为正式文献前的核验清单

  • [ ] 从论文、出版社、会议、ACL Anthology、OpenReview 或 arXiv 页面核对标题、作者、年份和版本。
  • [ ] 核对 DOI、venue、正式链接和是否存在撤稿/版本变化。
  • [ ] 阅读正文后补充方法、证据边界、结果和与现有研究的关系。

Semantic Scholar 记录用于发现候选,不是正式一手引文,也不支持对结果覆盖范围或研究质量的穷尽性判断。