Paper: arXiv 2510.15929
Authors: Lucas Eduardo Pereira Teles, Carlos M. S. Figueiredo
Abstract
This article presents a comparative study of large language models (LLMs) in the task of sentiment analysis of financial market news. This work aims to analyze the performance difference of these models in this important natural language processing task within the context of finance. LLM models are compared with classical approaches, allowing for the quantification of the benefits of each tested model or approach. Results show that large language models outperform classical models in the vast majority of cases.
Complexity vs Empirical Score
- Math Complexity: 3.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper involves practical data processing, model training, and performance evaluation on specific datasets, reflecting strong empirical rigor. Math complexity is moderate, focusing on applying existing techniques like TF-IDF and SVD rather than developing novel mathematical theory.
Research Flowchart
flowchart TD
A["Research Goal: Compare LLMs vs. Classical Models for Financial Sentiment Analysis"] --> B["Data: Financial Market News Dataset"]
B --> C["Methodology: Model Training & Evaluation"]
subgraph C ["Computational Process"]
C1["Classical Models e.g., SVM/LR"]
C2["LLMs e.g., BERT/FinBERT"]
end
C --> D["Performance Metrics: Accuracy, F1-Score"]
D --> E["Key Finding: LLMs consistently outperform classical models"]
E --> F["Outcome: Superior performance of LLMs for financial sentiment analysis"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper conducts a comparative study of large language models for sentiment analysis of financial market news.
Source sentence
This article presents a comparative study of large language models (LLMs) in the task of sentiment analysis of financial market news.
Assumes
No explicit assumptions found in the text.
Data
- Financial Phrase Bank (FPB) text-news public
Source sentence
Financial Phrase Bank (FPB): It contains data in both Portuguese and English (which was used). Each record is a general financial news item. In total, there are 4,840 records (but the one used in the project contains 5 more records, in this case 4,845 records)
- StockEmotions text-news public
Source sentence
StockEmotions: This dataset was built with the aim of detecting emotions in the financial market. It consists of 10,000 records, of which 8,000 are for training, 1,000 are for validation, and the remaining 1,000 are for testing.
- Tweet Financial News (TFN) text-news public
Source sentence
Tweet Financial News (TFN): The dataset contains financial news headlines collected in English. It is divided into training and validation, but only the validation data was used
StockTwits Twitter / X
Costs
Not applicable
Finds 2
- Large language models outperform classical models in most cases. performance
Source sentence
Results show that large language models outperform classical models in the vast majority of cases.
- Gemini model achieved the highest accuracy among tested models, maintaining accuracy above 70% across datasets. predictive-accuracy
Source sentence
Gemini was the most consistent model. Even though it did not have the highest accuracy among the three datasets, it still achieved good results compared to the other models, maintaining an accuracy above 70%.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 25 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.