Paper: arXiv 2412.10823
Authors: Yixuan Liang, Yuncong Liu, Neng Wang, Hongyang Yang, Boyu Zhang, Christina Dan Wang
Abstract
Financial sentiment analysis is crucial for understanding the influence of news on stock prices. Recently, large language models (LLMs) have been widely adopted for this purpose due to their advanced text analysis capabilities. However, these models often only consider the news content itself, ignoring its dissemination, which hampers accurate prediction of short-term stock movements. Additionally, current methods often lack sufficient contextual data and explicit instructions in their prompts, limiting LLMs’ ability to interpret news. In this paper, we propose a data-driven approach that enhances LLM-powered sentiment-based stock movement predictions by incorporating news dissemination breadth, contextual data, and explicit instructions. We cluster recent company-related news to assess its reach and influence, enriching prompts with more specific data and precise instructions. This data is used to construct an instruction tuning dataset to fine-tune an LLM for predicting short-term stock price movements. Our experimental results show that our approach improves prediction accuracy by 8% compared to existing methods.
Complexity vs Empirical Score
- Math Complexity: 3.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper’s methodology centers on data engineering (clustering, fine-tuning) and empirical evaluation with an 8% improvement over baselines, which is highly practical and implementation-heavy. The mathematics is relatively simple, relying on standard NLP techniques and statistical metrics rather than advanced theoretical derivations.
Research Flowchart
flowchart TD A["Research Goal<br>Predict Short-Term Stock Movements"] --> B["Input Data<br>Financial News & Company Data"] B --> C["Methodology: Dissemination & Context"] C --> D["Step 1: News Clustering & Reach Analysis"] C --> E["Step 2: Enrich Prompts with Context & Instructions"] D & E --> F["Step 3: Instruction Tuning Dataset Construction"] F --> G["Step 4: Fine-Tune LLM"] G --> H["Outcome: Enhanced Sentiment Prediction Model"] H --> I["Findings<br>8% Improvement in Accuracy"]
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 proposes a data‑driven clustering‑based method that incorporates news dissemination breadth, contextual data, and precise instructions, and uses an instruction‑tuning dataset to fine‑tune an LLM for short‑term stock movement prediction.
Source sentence
We propose a novel approach to enhance LLM-powered sentiment-based stock movement predictions by incorporating news dissemination breadth, contextual data, and precise instructions.
Assumes 2
- The centroid article of each news cluster encapsulates the most comprehensive information for LLM processing. data
Source sentence
i) The centroid article of each cluster encapsulates the most comprehensive information for LLM processing
- The size of a news cluster indicates the topic’s market impact, with larger clusters signifying more significant events. market
Source sentence
ii) The cluster size indicates the topic’s market impact, with larger clusters signifying more significant events
Data
- Finnhub API weekly financial news text-news public
Source sentence
We retrieve weekly financial news data, including titles and summaries, from the Finnhub API.
- Historical stock price and return data market-prices unknown
Source sentence
Our comprehensive dataset comprises 380 observations across 20 companies, spanning multiple market sectors
Costs
Not applicable
Finds 2
- The proposed approach improves prediction accuracy by 8% compared to existing methods. predictive-accuracy
Source sentence
Our experimental results show that our approach improves prediction accuracy by 8% compared to existing methods.
- Binary accuracy for stock movement prediction rises to 63% versus a 55% baseline. performance
Source sentence
Experimental results validate our approach, achieving 63% binary accuracy compared to the 55% baseline
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 20 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers 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.
Cited References (44)
- Blackboard Systems (1986)
- Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education (1983)
- Classification Problem Solving (1984)
- New Ways to Make Microcircuits Smaller (1980) — link
- New Ways to Make Microcircuits Smaller—Duplicate Entry (1980)
- Strategic explanations for a diagnostic consultation system (1984) — link
- Strategic Explanations in Consultation—Duplicate (1983)
- Poligon: A System for Parallel Problem Solving (1986)
and 36 more
- Transfer of Rule-Based Expertise through a Tutorial Dialogue (1979)
- The Engineering of Qualitative Models (2021)
- Attention Is All You Need (2017) — link
- Pluto: The ‘Other’ Red Planet (2015)
- Natural language based financial forecasting: a survey (2018)
- Bloomberggpt: A large language model for finance (2023) — link
- Automatic domain-specific sentiment lexicon generation with label propagation (2013)
- Big Data: Deep Learning for financial sentiment analysis (2018)
- Evaluation of sentiment analysis in finance: from lexicons to transformers (2020)
- FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019) — link
- Financial news predicts stock market volatility better than close price (2018)
- Sentiment analysis in financial texts (2017)
- Deep learning for financial sentiment analysis on finance news providers (2016)
- Instruct-fingpt: Financial sentiment analysis by instruction tuning of general-purpose large language models — → 2306.12659
- Fingpt: Open-source financial large language models — → 2306.06031
- Llama: Open and efficient foundation language models (2023) — link
- Enhancing TinyBERT for Financial Sentiment Analysis Using GPT-Augmented FinBERT Distillation (2024) — link
- Enhancing Financial Sentiment Analysis with Expert-Designed Hint (2024) — link
- Fine-Tuning Gemma-7B for Enhanced Sentiment Analysis of Financial News Headlines (2024) — link
- BERTopic-Driven Stock Market Predictions: Unraveling Sentiment Insights — → 2404.02053
- Large Language Model Adaptation for Financial Sentiment Analysis (2024) — link
- A Comparative Analysis of Fine-Tuned LLMs and Few-Shot Learning of LLMs for Financial Sentiment Analysis (2023) — link
- Fingpt: Instruction tuning benchmark for open-source large language models in financial datasets — → 2310.04793
- Enhancing financial sentiment analysis via retrieval augmented large language models — → 2310.04027
- Sentiment Analysis in Finance: From Transformers Back to eXplainable Lexicons (XLex) (2024)
- elahi2024combiningCombining Financial Data and News Articles for Stock Price Movement Prediction Using Large Language Models — → 2411.01368
- FinRobot: An Open-Source AI Agent Platform for Financial Applications using Large Language Models — → 2405.14767
- LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024) — link
- Inducing domain-specific sentiment lexicons from unlabeled corpora (2016)
- Stock movement prediction from tweets and historical prices (2018)
- Bert: Pre-training of deep bidirectional transformers for language understanding (2019)
- BERTopic: Neural topic modeling with a class-based TF-IDF procedure (2022) — link
- Rouge: A package for automatic evaluation of summaries (2004)
- Sentiment analysis in the era of large language models: A reality check (2023) — link
- FinRobot: AI Agent for Equity Research and Valuation with Large Language Models — → 2411.08804
- Enhancing Investment Analysis: Optimizing AI-Agent Collaboration in Financial Research — → 2411.04788
9 of 44 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.