Reproduce “Template-Based Financial Report Generation in Agentic and Decomposed Information Retrieval” 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
Yong-En Tian, Yu-Chien Tang, Kuang-Da Wang, An-Zi Yen, Wen-Chih Peng
Researchers at National Yang Ming Chiao Tung University compared two large language model approaches, AgenticIR and DecomposedIR, for automating template-based financial report generation from earnings call transcripts. DecomposedIR consistently outperformed AgenticIR, showing a 27% average improvement across evaluation metrics and maintaining its superiority in cross-domain validation.
Key insights
- DecomposedIR's question decomposition strategy is more effective for structured, template-based report generation than AgenticIR's multi-agent collaboration, especially for extracting and synthesizing financial data.
- Both AgenticIR and DecomposedIR benefited from self-reflection mechanisms, showing consistent performance improvements, but DecomposedIR maintained its performance advantage.
- A trade-off exists between report comprehensiveness and readability, as the more detailed reports generated by DecomposedIR sometimes resulted in higher complexity scores.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.