Reproduce “Beyond Binary Detection: Type-Aware Hallucination Analysis and Hierarchical Detection in Financial 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
Xiongyu Chen, Sizhen Zhu, Zhuoer Wang
The authors evaluate hallucination detection for financial question-answering over long filings by diagnosing errors by subtype and comparing a two-stage detector with an original run. The enhanced run raises binary accuracy from 80.2% to 95.6% and subtype accuracy from 44.7% to 56.1%. However, run counts differ (4,085 versus 4,086), and no hierarchy-only ablation establishes causation; fine-grained classification remains uneven.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.