FinSAgent: Corpus-Aligned Multi-Agent RAG Framework for Evidence-Grounded SEC Filing Question Answering
FinSAgent tries to make SEC-filing QA follow the filing corpus, not just the user prompt.
The paper argues that standard RAG and multi-agent systems miss evidence because their queries and rerankers lean too much on model priors and semantic similarity. FinSAgent assigns agents around the mandated 10-K item structure, decomposes queries with awareness of the local filing database, and uses multi-path retrieval with a feature-gated reranker. The authors report better retrieval coverage and answer correctness across five offline financial QA benchmarks, plus higher user ratings in a 1,000-rating online experiment. Source: ArXiv · AI/CL/LG's note.
The paper argues that standard RAG and multi-agent systems miss evidence because their queries and rerankers lean too much on model priors and semantic similarity. FinSAgent assigns agents around the mandated 10-K item structure, decomposes queries with awareness of the local filing database, and uses multi-path retrieval with a feature-gated reranker. The authors report better retrieval coverage and answer correctness across five offline financial QA benchmarks, plus higher user ratings in a 1,000-rating online experiment. Source: ArXiv · AI/CL/LG's note.
score 4