Tag
DiRAFT synthesizes financial transcripts by sampling concept-graph plans from aggregate discourse priors rather than individual source passages, and empirically evaluates residual confidentiality risks on two financial corpora, finding stochastic sampling best balances transition fidelity and privacy.
该论文提出了 EPSBench，一个用于评估 LLM 智能体季度盈利预测能力的基准，涵盖 86 家美国公司的 149 个公司季度数据，结果表明各模型的预测区间覆盖率显著低于标称的 80%。
该论文研究英文构建的 LLM 管道能否跨语言复用，通过支票提取、Mintaka 问答和 FinanceBench 三个任务评估准确性、推理努力与行为稳定性，结论是可有条件复用但仍需按语言按领域验证。
This paper presents a controlled comparison of single-agent versus multi-agent LLM setups on a financial document review task, finding that splitting work across three agents catches slightly more defects but increases false alarms and uses 2.2-2.4x the tokens, with simple self-sampling of the single agent often achieving better balanced accuracy.
研究人员发布了一个包含约16万篇巴西葡萄牙语金融新闻文章的大规模语料库，并基于语义相似度（余弦相似度、KNN、FAISS）建立了多分类弱监督基准，结果显示任务对标注质量和类别分布高度敏感。
FinGREAT 评估开放权重 LLM 在金融图引导检索与关系分类任务上的表现，利用 11 节点 11 边的复式记账图谱分析 1300 份 SEC 10-K 文件，并引入包含 483 段人工标注的关系分类基准，最强开源与闭源模型准确率分别为 84.1% 和 92.8%。
论文提出 MULTIFINGROUND 多语言金融基准，揭示语言模型虽能跨语言识别财务概念（准确率 97.3-99.4%），却在 26.6-41.9% 的情况下无法完整恢复对应的数值证据，凸显识别与证据落地之间的差距。
A Lancaster University study benchmarks summarisation methods on pre-trained transformers for long UK financial annual reports (FNS 2020 dataset), compares evaluation metrics, proposes the novel BRUGEscore (harmonic mean of ROUGE-2 and BERTscore), and validates results with statistical significance testing and adversarial corruption analysis.
The paper presents FinDialogLens, a hybrid LLM pipeline for event extraction over multi-party financial chatrooms to recover missed trades, achieving 92.1% and 94.3% accuracy on final price and trade outcome with GPT-4o while cutting LLM calls by 85% via a difficulty-aware router.
该研究评估了 Llama-3 8B、Qwen-2.5 14B 和 Llama-3.3 70B 等 LLM 在从 SEC 10-K 文件中提取财务数据时的表现，发现将模型规模与文档结构复杂度对齐可在零样本设置下获得高准确率，从而缓解量化金融数据集中超过 70% 公司面临的缺失数据偏差问题。
This paper proposes PALM, a low-rank adapter method to adapt financial language models to specific time points without full retraining, reducing look-ahead bias and computational costs in financial backtests.
This paper tests the assumption that sentiment tools validated against human labels also predict market signals in financial NLP, finding that benchmark agreement does not determine predictive effectiveness and spam influences volume-based analysis.
This paper audits temporal leakage in financial news NLP benchmarks across multiple models, finding that random splits inflate performance metrics and identifies M&A events as a category with a localized positive signal under chronological evaluation.
CLAIR-Fin is an adversarial multi-agent framework designed to enhance faithfulness in cross-modal financial question answering by decomposing questions into atomic claims and using adaptive debate and verification to reduce hallucinations.
This paper presents a method using Group Relative Policy Optimization (GRPO) to fine-tune an open-weight language model for generating actionable financial advice, outperforming commercial LLMs under a judge-independent CATE evaluation while also matching safety criteria.
This paper introduces LabelFusion-TS, which fuses a fine-tuned RoBERTa encoder, a prompted LLM, and time-series transformers over market data to classify Federal Reserve communication as hawkish, dovish, or neutral. The fused system achieves 70.2% weighted F1, outperforming a zero-shot LLM and showing early evidence that market time series help financial text classification.
This paper presents AWARE-FX, an auditable AI/NLP decision-support system that extracts and scores corporate foreign-exchange hedging disclosures from annual reports, evaluated on 24,909 Hong Kong firm-years with FinBERT, ModernBERT, and Qwen3-8B comparisons.
This paper presents team HSA_CORAL's submission to the FinCausal 2026 shared task, comparing encoder-only, encoder-decoder, and decoder-only LLMs for extractive question answering of cause-effect relations in financial narratives. Fine-tuned GPT-4.1 Mini achieved top scores in the English subtask and third in Spanish.
Presents a framework for financial sentiment analysis using distillation with synthetic data, transferring knowledge from a large teacher to compact student models, with clustering-based seed selection for efficient low-resource domain adaptation.
This paper introduces Semantic State Abstraction Interfaces (SSAI) to separate representation hypotheses from optimization variance in LLM-augmented portfolio decisions. It concludes that SSAI's apparent advantage is largely a basket-selection effect, with dense encodings and principal components performing better empirically.