NeSy-RAG: Neuro-Symbolic RAG for Explainable Question Answering
The paper turns retrieved text into Prolog so the answer can be traced step by step.
NeSy-RAG generates Boolean predicates from retrieved chunks, retrieves and composes them into Prolog queries, then executes those queries for deterministic answers. It also detects when missing user facts could change the result and prompts for that context. On ShARC, the authors report 61.1% accuracy without domain-specific training, versus 42.8% for a same-model RAG baseline. ArXiv · AI/CL/LG's note
NeSy-RAG generates Boolean predicates from retrieved chunks, retrieves and composes them into Prolog queries, then executes those queries for deterministic answers. It also detects when missing user facts could change the result and prompts for that context. On ShARC, the authors report 61.1% accuracy without domain-specific training, versus 42.8% for a same-model RAG baseline. ArXiv · AI/CL/LG's note
score 4