Presenter(s)
Trevor Nindl
Abstract
This study investigated how news sentiment analysis improves stock price prediction accuracy. Two Long Short-Term Memory (LSTM) neural networks were created using PyTorch, a baseline model without sentiment analysis trained using 22 technical features, and another model with sentiment analysis scores from Alpha Vantage’s sentiment analysis API. The historical data gathered spans from 2018-2025 and it covers a total of seven securities at 20-minute intervals. There are three volatile stocks (Tesla, NVIDIA, AMC), three stable stocks (Johnson & Johnson, Coca-Cola, Procter & Gamble), and a single S&P 500 index (SPY). The models were run in the same testing environment that mimicked a real-world trading platform using historical stock data. They were both also implemented on two identical Alpaca live paper trading accounts where they traded autonomously. Performance was evaluated using total returns, win rate, and statistical significance. A real-world environment like this gave us evidence that sentiment analysis inclusion can be worth it when it comes to stock price predictions in certain situations.
College
College of Liberal Arts
Department
Computer Science
Campus
Winona
First Advisor/Mentor
Collin Engstrom
Second Advisor/Mentor
Trung Nguyen
Third Advisor/Mentor
Mingrui Zhang
Location
Kryzsko Great River Ballroom, Winona, Minnesota; United States
Start Date
4-23-2026 1:00 PM
End Date
4-23-2026 2:00 PM
Presentation Type
Poster Session
Format of Presentation or Performance
In-Person
Session
2a=1pm-2pm
Poster Number
43
Impact of News Sentiment Analysis on the Performance of Stock Trading Models
Kryzsko Great River Ballroom, Winona, Minnesota; United States
This study investigated how news sentiment analysis improves stock price prediction accuracy. Two Long Short-Term Memory (LSTM) neural networks were created using PyTorch, a baseline model without sentiment analysis trained using 22 technical features, and another model with sentiment analysis scores from Alpha Vantage’s sentiment analysis API. The historical data gathered spans from 2018-2025 and it covers a total of seven securities at 20-minute intervals. There are three volatile stocks (Tesla, NVIDIA, AMC), three stable stocks (Johnson & Johnson, Coca-Cola, Procter & Gamble), and a single S&P 500 index (SPY). The models were run in the same testing environment that mimicked a real-world trading platform using historical stock data. They were both also implemented on two identical Alpaca live paper trading accounts where they traded autonomously. Performance was evaluated using total returns, win rate, and statistical significance. A real-world environment like this gave us evidence that sentiment analysis inclusion can be worth it when it comes to stock price predictions in certain situations.
Comments
Nindl, Trevor