Reproduce “Repair Before Reinforce: Context-Augmented Knowledge Graph Reasoning for Multi-Hop Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tharaka D. Fonseka, Niraj K. Jha
Authors train biomedical question-answering models by pairing target knowledge-graph facts with related triples from the same source-text chunk. They audit and repair one-hop failures before reinforcement learning. Context-trained models outperform fact-only models on three-, four-, and five-hop questions. On Diabetes context-graph-grounded four-hop questions, repaired-initialization reinforcement learning reaches 97.51% accuracy versus 94.35% from unrepaired initialization in synthetic evaluations of two disease-specific knowledge graphs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.