Paper: arXiv 2502.05186

Authors: Furkan Karadaş, Bahaeddin Eravcı, Ahmet Murat Özbayoğlu

Abstract

In an era where financial markets are heavily influenced by many static and dynamic factors, it has become increasingly critical to carefully integrate diverse data sources with machine learning for accurate stock price prediction. This paper explores a multimodal machine learning approach for stock price prediction by combining data from diverse sources, including traditional financial metrics, tweets, and news articles. We capture real-time market dynamics and investor mood through sentiment analysis on these textual data using both ChatGPT-4o and FinBERT models. We look at how these integrated data streams augment predictions made with a standard Long Short-Term Memory (LSTM model) to illustrate the extent of performance gains. Our study’s results indicate that incorporating the mentioned data sources considerably increases the forecast effectiveness of the reference model by up to 5%. We also provide insights into the individual and combined predictive capacities of these modalities, highlighting the substantial impact of incorporating sentiment analysis from tweets and news articles. This research offers a systematic and effective framework for applying multimodal data analytics techniques in financial time series forecasting that provides a new view for investors to leverage data for decision-making.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper employs standard deep learning architectures (LSTM, FinBERT) without advanced mathematical derivations, but demonstrates rigorous empirical testing with diverse real-world data sources, multiple models, and clear performance metrics.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Accurate Stock Price Prediction"] --> B["Data Collection & Preprocessing"]
  B --> C{"Multimodal Data Sources"}
  C --> D["Traditional Financial Metrics"]
  C --> E["Social Media Tweets"]
  C --> F["News Articles"]
  
  D --> G["Computational Processing"]
  E --> G
  F --> G
  
  subgraph G ["LSTM Model Pipeline"]
      G1["Sentiment Analysis<br>ChatGPT-4o & FinBERT"] --> G2["Feature Integration"]
      G2 --> G3["Long Short-Term Memory<br>LSTM Network"]
      G3 --> G4["Price Prediction Output"]
  end
  
  G4 --> H["Key Findings & Outcomes"]
  
  H --> I["5% Performance Gain<br>vs. Baseline LSTM"]
  H --> J["Senti-mental Analysis<br>Significantly Enhances Accuracy"]
  H --> K["Effective Framework for<br>Multimodal Financial Analytics"]

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 multimodal machine learning approach that combines traditional financial metrics with sentiment analysis from tweets and news articles to improve stock price prediction.

Source sentence
a multimodal machine learning approach for stock price prediction by combining data from diverse sources, including traditional financial metrics, tweets, and news articles.

Assumes 5

  • Missing data are filled using forward‑filling techniques to preserve dataset continuity. data
    Source sentence
    Using forward-filling techniques, we effectively imputed missing values while preserving the dataset's integrity.
  • Hyperparameters for tweet interaction weights are set to 0.3 without further optimization. model
    Source sentence
    In this study, we set these parameters to 0.3 as an initial estimate without any optimization.
  • Training uses data from 2018‑2022 and testing on 2023, assuming this temporal split reflects realistic out‑of‑sample performance. execution
    Source sentence
    Our model was trained using data from 2018 to 2022, while data from 2023 was reserved for testing.
  • Tweets must have at least 100 likes to be included, assuming higher‑engagement tweets are more informative. data
    Source sentence
    We required each tweet to have at least 100 likes to prior…
Show 1 more
  • Sentiment scores from FinBERT and ChatGPT‑4o are reliable indicators of market sentiment. model
    Source sentence
    FinBERT is an adaptation of the BERT architecture and is specifically trained to process financial texts

Data

  • Yahoo Finance price data · Yahoo Finance market-prices public
    2018-2023 · Walmart Inc., Walt Disney Co., Microsoft
    Source sentence
    We collected comprehensive financial data for each company from Yahoo Finance, including historical price movements and other relevant financial indicators. The data gathered for each stock ranges from 2018 to 2023.
  • Twitter tweets · Twitter alternative public
    Walmart, Disney, Microsoft related tweets
    Source sentence
    We collected tweets relevant to each stock to understand public opinion on market dynamics by filtering for specific keywords. Keywords included the company name or stock ticker symbol, such as "Microsoft or MSFT." We required each tweet to have at least 100 likes
  • News articles text-news unknown
    Source sentence
    real‑time market dynamics and investor mood through sentiment analysis on these textual data using both ChatGPT‑4o and FinBERT models.

Recognised sources: Twitter / X Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Incorporating sentiment features improves prediction performance for Walt Disney Co.; weighted tweet sentiment yields R² = 0.9086, MAE = 0.0150 versus price‑only R² = 0.8915, MAE = 0.0166. performance
    Source sentence
    the model that combined price data with the weighted sentiment score of tweets achieved higher R² scores (R² = 0.9086, MAE = 0.0150) than just using price data (R² = 0.8915, MAE = 0.0166)
  • For Microsoft, combining price data with tweets, news, and technical indicators yields R² = 0.9640, MAE = 0.0193, outperforming price‑only R² = 0.9497, MAE = 0.0233. performance
    Source sentence
    the model with combined data achieved better results with tweets, news articles, and technical indicators (R² = 0.9640, MAE = 0.0193) compared to using only price data (R² = 0.9497, MAE = 0.0233)
  • For Walmart, the best result is achieved with price data plus RSI and SMA, giving R² = 0.9282 and MAE = 0.0181. performance
    Source sentence
    For Walmart, the combination of price data with RSI and SMA yielded the best results (R² = 0.9282, MAE = 0.0181)
  • Overall, incorporating multimodal data increases forecast effectiveness of the reference LSTM model by up to 5%. predictive-accuracy
    Source sentence
    Our study's results indicate that incorporating the mentioned data sources considerably increases the forecast effectiveness of the reference model by up to 5%.
  • The multimodal approach demonstrates that sentiment analysis from textual data improves model performance compared to using price data alone. performance
    Source sentence
    The results revealed that incorporating sentiment analysis of textual data (tweets and news articles) into the model improved its performance compared to using price data alone.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 40 related papers). Coverage is partial, so this is neither replication nor refutation.

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.