Research map / Large language models / Language, safety and interpretability
Interpreting Fedspeak with Confidence: A LLM-Based Uncertainty-Aware Framework Guided by Monetary Policy Transmission Paths
Rui Yao, Qi Chai, Jinhai Yao, Siyuan Li, Junhao Chen, Qi Zhang, Hao Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add dynamic uncertainty decoding (PU) (ablation) | fine-tuned Qwen3-14B with augmentation · FOMC (all categories) | Weighted F1: 0.7378 -> 0.7426 | Improved |
| add monetary policy transmission path information (ablation) | Qwen3-14B framework · FOMC (all categories) | Weighted F1: 0.6699 -> 0.7426 | Improved |
| add financial entity relationships (ablation) | Qwen3-14B framework · FOMC (all categories) | Weighted F1: 0.6551 -> 0.7426 | Improved |
| add all proposed modules to fine-tuned Qwen3 (ablation) | Qwen3-14B · FOMC (all categories) | Macro F1: 0.6360 -> 0.7327 | Improved |
| add rules to the augmentation prompt (ablation) | Gemini-2.5-Pro augmentation prompt · FOMC (augmentation prompt, Gemini-2.5-Pro) | Weighted F1: 0.5073 -> 0.6695 | Improved |
| add examples to the augmentation prompt (ablation) | Gemini-2.5-Pro augmentation prompt · FOMC (augmentation prompt, Gemini-2.5-Pro) | Weighted F1: 0.5024 -> 0.6695 | Improved |
| use text-based stance labels instead of numeric labels (ablation) | fine-tuned Qwen3-14B · FOMC (all categories) | Weighted F1: 0.7114 -> 0.7378 | Improved |
| fine-tune with LoRA on FOMC data | GLM-4-9B · FOMC (all categories) | Macro F1: 0.3152 -> 0.6390 | Improved |
| cluster label tokens before uncertainty computation (ablation) | uncertainty decoding · FOMC (all categories) | Weighted F1: 0.7404 -> 0.7426 | Improved |
| apply framework to Phi-4 instead of Qwen-3 | Phi-4 · FOMC (transfer to Phi-4) | Weighted F1: 0.7225 -> 0.7241 | No clear effect |
| use conservative/greedy-k variants instead of greedy-c with candidate sampling (ablation) | uncertainty decoding · FOMC (all categories) | Weighted F1: 0.7426 -> 0.7364 | Hurt |
| use Ours instead of GPT-4.1 | press conference transcripts · FOMC (press conference) | Weighted F1: 0.6900 -> 0.6699 | Hurt |
Closest papers by meaning
- Words That Unite The World: A Unified Framework for Deciphering Central Bank Communications (NeurIPS 2025 · similarity 0.86)
- FinBen: A Holistic Financial Benchmark for Large Language Models (NeurIPS 2024 · similarity 0.82)
- Integrating Local and Global Entropy for Uncertainty Quantification in LLMs (NeurIPS 2026 · similarity 0.82)
- STEER-ME: Assessing the Microeconomic Reasoning of Large Language Models (NeurIPS 2025 · similarity 0.82)
- Distinguishing the Knowable from the Unknowable with Language Models (ICML 2024 · similarity 0.82)
- FinSearchComp: Towards a Realistic, Expert-Level Evaluation of Financial Search and Reasoning (ICLR 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
