IGMiRAG Improves RAG Efficiency And Effectiveness

Researchers (Xingliang Hou) on Feb 7, 2026 propose IGMiRAG, a retrieval-augmented generation framework that builds a hierarchical heterogeneous hypergraph with deductive pathways, a question parser for intuition-guided retrieval, dual-focus retrieval anchors, and a bidirectional diffusion algorithm. Evaluations report 4.8% exact-match and 5.0% F1 improvements over a state-of-the-art baseline, with token costs adaptive (average 6.3k+, minimum 3.0k+).