Financial QA
QA: Question Answering
Overview
Financial question answering studies how to answer questions about financial filings, statements, spreadsheets, charts, earnings calls, and regulations, where errors in facts or calculations can carry significant costs. Common approaches combine retrieval-augmented generation—using methods such as BM25, dense retrieval, hybrid search, and reranking—with table-aware parsing, structured data tools, or executable code for numerical reasoning. Benchmarks test tasks including multi-step calculations, evidence citation, multilingual and multimodal understanding, and retrieval from long documents; newer systems also use claim-level verification, abstention, and selective checks. A central challenge is that finding relevant evidence does not ensure correct reasoning, while stronger verification and retrieval can add cost or sometimes overturn sound answers.
Latest papers
51 papers5 papers in the last four weeks, level with the four weeks before. 0.0% of all new papers.