Reproduce “Multi-hop Reading Comprehension via Deep Reinforcement Learning based Document Traversal” 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

Alex Long, Joel Mason, Alan Blair, Wei Wang

The authors introduce a multi-hop reading system that selects connected sentences sequentially before an answer reader extracts answers. A linker builds a sparse sentence graph; a reinforcement-learning policy chooses next sentences using the question and visited evidence. On WikiHop, it scores 65.12 exact-match accuracy versus 44.06 for the baseline; a fully connected graph scores 33.45, supporting graph design, not sequential choice alone.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.