---
title: "Beyond Sentiment: Structured Information Extraction from Financial News"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.28496v1"
permalink: "https://feed.astralor.com/items/itm_a22c7e313fed28868e23e029"
published: "2026-07-30T16:41:55.000Z"
score: 73
channel: "finance"
tags: ["FinBERT","FNSPID","LLaMA-3.1-70B","Structured Information Extraction","Financial NLP"]
---

# Beyond Sentiment: Structured Information Extraction from Financial News

## AI commentary

> This paper's value lies in efficiently proving the old intuition that 'sentiment scores are not enough' with a reproducible framework and clear quantitative evidence. For those working on financial sentiment or stock prediction, the 53.5% disagreement rate and 0.600 combined F1 are directly useful baselines. Note that LLM-based structured extraction costs far more than FinBERT, so real-world adoption requires a cost-benefit tradeoff.

## At a glance

Financial sentiment analysis has become standard for news-driven stock prediction, but it reduces rich news to a single polarity score. Researchers propose a structured information extraction framework using LLaMA-3.1-70B to extract six semantic dimensions from 41,618 news-stock pairs in the FNSPID dataset. Experiments show FinBERT sentiment features perform well under nonlinear models (F1=0.576) but weakly under linear models (F1=0.230). LLM-extracted structured features show a 53.5% systematic disagreement with sentiment, and combining both yields F1=0.600, significantly better than either alone. Ablation confirms non-sentiment dimensions independently add ΔF1=+0.019 beyond FinBERT. The results demonstrate sentiment-semantics decoupling is systematic and exploitable, opening new directions for multi-dimensional financial NLP.

## Key points

- Proposes a structured information extraction framework using LLaMA-3.1-70B to extract six semantic dimensions from financial news, going beyond a single sentiment score.
- On 41,618 news-stock pairs, FinBERT sentiment features show strong performance under nonlinear models (F1=0.576) but weak performance under linear models (F1=0.230), revealing a highly nonlinear sentiment-return relationship.
- Structured features and sentiment features exhibit a 53.5% systematic disagreement rate, and combining both yields F1=0.600, significantly outperforming either alone (p<0.0001) with consistent gains across all seven event types.
- Feature importance analysis shows balanced contributions from all six dimensions (14-21%), demonstrating that compressing news into a single sentiment score incurs substantial information loss.

## Full text

Financial sentiment analysis has become a standard component in news-driven stock prediction, yet it reduces rich, multi-dimensional news articles to a single polarity score. We hypothesize that financial news encodes multiple orthogonal information dimensions---event type, impact scope, temporal horizon, and semantic confidence---that sentiment alone cannot capture, and that these dimensions carry independent predictive value. To test this hypothesis, we propose a structured information extraction framework that leverages LLaMA-3.1-70B to extract six semantic dimensions from financial news. Through large-scale experiments on 41,618 news--stock pairs from the FNSPID dataset, we find that (i) FinBERT sentiment features exhibit strong predictive power under nonlinear models (F1=0.576) but substantially weaker performance under linear models (F1=0.230), revealing a highly nonlinear sentiment--return relationship; (ii) LLM-extracted structured features, while individually weaker, capture information orthogonal to sentiment, as evidenced by a 53.5% systematic disagreement rate between the two approaches; and (iii) combining both signal sources yields F1=0.600, significantly outperforming either alone ($p < 0.0001$), with consistent improvements across all seven event types. Ablation experiments confirm that non-sentiment structural dimensions (event type, impact subject, time horizon, confidence) independently contribute $Δ\text{F1} = +0.019$ beyond FinBERT alone. Feature importance analysis reveals balanced contributions from all six extracted dimensions (14--21%), demonstrating that compressing news into a single sentiment score incurs substantial information loss. Our results suggest that the sentiment--semantics decoupling in financial text is systematic and exploitable, opening a new direction for multi-dimensional financial NLP.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_a22c7e313fed28868e23e029