DEEP-PolyU/LinearRAG
[ICLR 2026] LinearRAG: Linear Graph Retrieval Augmented Generation on Large-scale Corpora
What it solves
LinearRAG 解決了傳統 GraphRAG 系統的高成本與緩慢速度問題。具體而言,它消除了在圖建構階段通常需要使用 LLM 來抽取實體之間關係的昂貴代幣消耗。
How it works
LinearRAG 並不建立顯式的關係圖,而是採用「無關係」的建構方式。它結合輕量級的實體辨識與語意連結,生成圖結構。這使系統能在單次檢索過程中透過語意橋接執行多跳推理與深度檢索,並維持線性時間與空間複雜度。
Who it’s for
此專案適合在大規模語料庫上使用 Retrieval‑Augmented Generation(RAG)的開發者與研究者,這些使用者需要圖形檢索的推理能力,但又不想承擔 LLM 驅動圖建構的高昂計算成本。
Highlights
- Zero LLM token cost: Eliminates LLM usage during the graph building process.
- Linear Scalability: Operates with linear time and space complexity, making it faster and easier to scale to large datasets.
- Multi-hop Reasoning: Supports complex reasoning across different pieces of information without needing explicit relational graphs.
- Context-Preserving: Uses semantic linking to ensure comprehensive contextual understanding.