Stock Price Prediction Using LSTM Networks and Sentiment Analysis

from Financial News

Akshay Arjun Shinde
Department of Data Science, Kirti College, Dadar, Mumbai, India
Akshay29shinde@[Link]
Guide - [Link] Margaj

Abstract
Stock market prediction is a complex and challenging task due to the highly volatile,
nonlinear, and dynamic nature of financial markets. Traditional statistical models often fail to
capture long-term dependencies and external market influences that affect stock price
movements. In recent years, deep learning techniques, particularly Long Short-Term Memory
(LSTM) networks, have shown strong performance in time-series forecasting problems. In
addition, financial news plays a significant role in influencing investor sentiment and market
behavior.
This research paper proposes a hybrid approach that combines LSTM-based time-series
modeling with sentiment analysis of financial news to improve stock price prediction
accuracy. Historical stock price data is used to learn temporal patterns through LSTM
networks, while sentiment scores extracted from financial news articles provide qualitative
insights into market emotions. The sentiment information is integrated with numerical stock
features to enhance the predictive capability of the model. Experimental analysis
demonstrates that the proposed hybrid model outperforms models based solely on historical
price data. The results indicate that incorporating sentiment analysis leads to improved
prediction accuracy and better responsiveness to market-moving events.

Keywords: Stock Price Prediction, LSTM Networks, Sentiment Analysis, Financial News,
Deep Learning, Time Series Forecasting

1. Introduction

Stock market prediction has long been an area of interest for researchers, economists, traders,
and financial institutions. Accurate prediction of stock prices can lead to better investment
strategies, risk mitigation, and improved financial planning. However, stock markets are
inherently complex systems influenced by a wide range of factors including historical price
movements, company fundamentals, macroeconomic indicators, geopolitical events, and
investor psychology. This complexity makes stock price forecasting a difficult and uncertain
task. … Fama (1970)

Traditional stock price prediction techniques include statistical models such as linear
regression, autoregressive integrated moving average (ARIMA), and generalized
autoregressive conditional heteroskedasticity (GARCH). While these models have been
widely used, they assume linearity and stationarity in data, which often does not hold true in
real-world financial markets. As a result, their predictive performance is limited, especially
during periods of high volatility.
With the advancement of machine learning and deep learning, data-driven approaches have
gained prominence in financial forecasting. Deep learning models can automatically learn
complex patterns from large datasets without explicit feature engineering. Among these
models, Recurrent Neural Networks (RNNs) are particularly suitable for time-series data.
However, standard RNNs suffer from issues such as vanishing and exploding gradients,
which limit their ability to capture long-term dependencies.

Long Short-Term Memory (LSTM) networks were introduced to overcome these limitations.
LSTM networks include memory cells and gating mechanisms that allow them to retain
relevant information over long sequences. This makes them highly effective for modeling
stock price movements, which often depend on long-term historical trends. …Ticknor (2013)

In addition to historical price data, qualitative information such as financial news

significantly impacts stock markets. News related to earnings reports, mergers, policy
changes, or economic indicators can cause sudden price fluctuations. Investor reactions to
such news are driven by sentiment, which can be positive, negative, or neutral. Ignoring this
information can lead to incomplete or inaccurate predictions. Therefore, integrating sentiment
analysis with time-series forecasting models has the potential to improve stock price
prediction accuracy. This research aims to develop such an integrated framework using
LSTM networks and financial news sentiment analysis. ... Bollen et al. (2011)

2. Literature Review
Numerous studies have explored stock price prediction using various analytical techniques.
Early research primarily relied on statistical and econometric models. Methods such as
ARIMA and exponential smoothing were widely used due to their simplicity and
interpretability. However, these approaches require strong assumptions about data
distribution and often fail to handle nonlinear patterns. ... Fama (1970)

Machine learning techniques marked a significant shift in financial forecasting research.

Support Vector Machines (SVM), decision trees, random forests, and k-nearest neighbors
have been applied to predict stock prices and trends. These models showed improved
performance over traditional statistical methods, particularly in capturing nonlinear
relationships. Nevertheless, most machine learning models require extensive feature
engineering and struggle with sequential dependencies in time-series data.
…Ticknor (2013)

Deep learning models, especially neural networks, have gained attention due to their ability
to automatically extract features from raw data. Artificial Neural Networks (ANNs) were
among the first deep learning models applied to financial prediction. However, ANNs treat
input data as independent observations and do not explicitly model temporal relationships.

Recurrent Neural Networks address this limitation by incorporating feedback connections,

enabling them to process sequential data. Despite their advantages, RNNs encounter training
difficulties when dealing with long sequences. LSTM networks were introduced to solve this
problem and have since been widely used in stock price prediction tasks.
…Hochreiter & Schmidhuber (1997)
Parallel to price-based prediction methods, sentiment analysis has emerged as an important
research area in financial analytics. Sentiment analysis uses natural language processing
(NLP) techniques to extract emotional tone from textual data such as news articles, financial
reports, and social media posts. Several studies have shown a strong correlation between
market sentiment and stock price movements. Positive news tends to drive prices upward,
while negative news often results in price declines.

Recent research has explored hybrid models that combine numerical and textual data. These
studies demonstrate that integrating sentiment analysis with deep learning models can
enhance prediction accuracy. However, challenges remain in effectively aligning news
sentiment with price data and selecting appropriate model architectures. This research builds
upon existing work by proposing a structured hybrid LSTM framework that incorporates
financial news sentiment in a systematic manner.

3. Research Methodology
This document has been set up with a set of styles that should give the correct layout
automatically. For example, the heading above is in the style ‘Paper heading - 2’. Please
always use these styles. They include the spacing required between different items
automatically; it should not be necessary to add blank lines.
Styles are included in this template for the following elements of a paper: Title, Author,
Author address, Abstract, Headings, Figure captions, Table titles, Equations, References.

3.1 Data Collection

The proposed model utilizes two main types of data: historical stock price data and financial
news data. Historical stock data includes daily values such as opening price, closing price,
highest price, lowest price, and trading volume. This data provides quantitative insights into
market behavior over time.
Financial news data consists of articles related to selected stocks and overall market
conditions. News articles are collected from reliable financial news sources and are time-
stamped to ensure proper alignment with stock price data. Only relevant news items are
considered to avoid noise and irrelevant information. … Liu (2012)
3.2 Data Preprocessing
Preprocessing is a crucial step to ensure data quality and model performance. Stock price data
is checked for missing values and outliers. Missing values are handled using appropriate
interpolation or removal techniques. The data is normalized using Min-Max scaling to bring
all features within a uniform range, which helps accelerate model training.
For news data, text preprocessing is performed using NLP techniques. This includes
converting text to lowercase, removing punctuation and stop words, tokenization, and
lemmatization. These steps reduce noise and improve the quality of sentiment analysis.
… Kim (2014)
3.3 Sentiment Analysis
Sentiment analysis is applied to the cleaned news articles to determine the emotional tone of
the text. Each article is classified into positive, negative, or neutral sentiment categories. A
sentiment score is assigned to each article based on polarity. To align sentiment with stock
prices, daily sentiment scores are calculated by aggregating sentiment values of all news
articles published on the same day.
These sentiment scores serve as external indicators of market mood and are used as additional
input features in the prediction model. … Devlin et al. (2019)
3.4 LSTM Network Architecture
The LSTM model is designed to process sequential data and capture long-term dependencies.
The input layer receives a combination of historical stock price features and sentiment scores.
One or more LSTM layers are used to model temporal relationships. Dropout layers are
included to prevent overfitting. Finally, a dense output layer generates the predicted stock
price.
The LSTM network learns both price trends and sentiment-driven market reactions, enabling
more accurate predictions.

3.5 Model Training and Evaluation

The dataset is divided into training, validation, and testing subsets. The model is trained using
the training set, while the validation set is used to tune hyperparameters and prevent
overfitting. The final evaluation is performed on the test set.
Performance is measured using evaluation metrics such as Mean Absolute Error (MAE),
Mean Squared Error (MSE), and Root Mean Square Error (RMSE). These metrics provide a
quantitative assessment of prediction accuracy.

4. Equations, Figures and Tables

The LSTM model operates based on memory cells and gating mechanisms. The general
representation of an LSTM unit is given by: ... Hochreiter & Schmidhuber (1997)

Equation 1

where represents the input vector consisting of stock prices and sentiment features at time
, and is the hidden state.

Fig. 1: Conceptual architecture of the hybrid LSTM and sentiment analysis model.

Day Open Close Volume Sentiment Score

1 120 122 1.5M 0.72
2 122 119 1.8M -0.45

Table 1: Sample representation of stock price data with sentiment scores

 5. Results and Discussion

The experimental results indicate that the LSTM model incorporating sentiment analysis
outperforms the baseline LSTM model that relies only on historical stock price data. The
hybrid model demonstrates lower error values across all evaluation metrics. This
improvement highlights the importance of including external information in financial
forecasting. … Fischer & Krauss (2018)

The model shows enhanced responsiveness to sudden market changes caused by significant
news events. For instance, negative news related to economic uncertainty results in lower
predicted prices, aligning closely with actual market behavior. Similarly, positive sentiment
from earnings announcements leads to improved prediction accuracy.

….. Kearney & Liu (2014)

While the proposed model performs well, certain limitations remain. The quality of sentiment
analysis depends heavily on the accuracy of text classification. Additionally, delays between
news publication and market reaction can introduce noise. Despite these challenges, the
results confirm that sentiment analysis adds meaningful value to stock price prediction.

6. Conclusion

This research presents a comprehensive hybrid framework for stock price prediction that
combines LSTM networks with sentiment analysis derived from financial news. By
integrating quantitative historical data with qualitative sentiment information, the proposed
model achieves improved prediction accuracy and robustness.

The findings demonstrate that financial news sentiment plays a significant role in influencing
stock prices and should not be ignored in predictive modeling. LSTM networks effectively
capture long-term dependencies in time-series data, making them suitable for financial
forecasting tasks. Future research may explore advanced natural language processing models
such as transformer-based architectures, real-time prediction systems, and portfolio-level
forecasting approaches. ….. Devlin et al. (2019)
References
1. Fischer, T., & Krauss, C. (2018). Deep learning with long short-term memory networks for financial market
predictions. European Journal of Operational Research, 270(2), 654–669
2. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780.
3. Pang, B., & Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information
Retrieval, 2(1–2), 1–135.
4. Bollen, J., Mao, H., & Zeng, X. (2011). Twitter mood predicts the stock market. Journal of Computational
Science, 2(1), 1–8.
5. Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers
for language understanding. Proceedings of NAACL-HLT, 4171–4186.
6. Liu, B. (2012). Sentiment analysis and opinion mining. Morgan & Claypool Publishers.
7. Kim, Y. (2014). Convolutional neural networks for sentence classification. Proceedings of the 2014 Conference on
Empirical Methods in Natural Language Processing, 1746–1751.
8. Ticknor, J. L. (2013). A Bayesian regularized artificial neural network for stock market forecasting. Expert
Systems with Applications, 40(14), 5501–5506.
9. Zhang, X., Zhao, J., & LeCun, Y. (2015). Character-level convolutional networks for text classification. Advances
in Neural Information Processing Systems, 28, 649–657.
10. Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. Journal of Finance, 25(2),
383–417.
11. Schumaker, R. P., & Chen, H. (2009). Textual analysis of stock market prediction using breaking financial news.
ACM Transactions on Information Systems, 27(2), 1–19.
12. Kearney, C., & Liu, S. (2014). Textual sentiment in finance: A survey of methods and models. International
Review of Financial Analysis, 33, 171–185.