📈 Stock Price Prediction using Twitter Sentiment (Hybrid RoBERTa-GRU)

A comparative study and implementation of Hybrid Neural Architectures for stock market trend prediction. This project proposes a novel RoBERTa + GRU hybrid model that analyzes Twitter sentiment to forecast stock price movements, achieving state-of-the-art performance compared to baseline BERT models.

Research Highlight: The proposed Hybrid GRU model achieved a classification accuracy of 60.8%, outperforming standard BERT (56.1%) and RoBERTa (57.3%) baselines on the test dataset.

🚀 Key Features

• Hybrid Architecture: Combines RoBERTa (for contextual embedding) with a Bi-Directional GRU (for sequential modeling) to capture long-range dependencies in financial text.
• Large-Scale Analysis: Trained on a dataset of 300,000+ tweets (2015–2020) covering major tech stocks (Apple, Google, Microsoft, Tesla).
• Robust Preprocessing: Implements custom regex pipelines to handle noisy social media text (removing tickers, URLs, and bot spam).
• Comprehensive Evaluation: Benchmarked against 4 different architectures:
  1. BERT Base
  2. RoBERTa Base
  3. Modified RoBERTa (Custom Classification Head)
  4. Hybrid RoBERTa-GRU (Winner)

🛠️ Tech Stack

• Language: Python 3.8+
• Deep Learning: PyTorch
• NLP: Hugging Face Transformers, NLTK
• Data Processing: Pandas, NumPy, Scikit-learn
• Data Source: Kaggle Tweet Dataset & Yahoo Finance API (yfinance)

📊 Performance Metrics

The models were evaluated on the held-out test set using Accuracy, Precision, Recall, and F1-Score. The Hybrid RoBERTa-GRU demonstrated superior performance across all metrics, indicating better handling of sequential context in financial tweets.

Model	Accuracy	Precision	Recall	F1 Score
Hybrid RoBERTa-GRU (Proposed)	0.608	0.596	0.589	0.595
Modified RoBERTa	0.585	0.569	0.573	0.577
RoBERTa Base	0.573	0.567	0.576	0.566
BERT Base	0.561	0.568	0.555	0.547

📂 Project Structure

├── stock_classifier_tweet.py     # Main model training & evaluation loop (PyTorch)
├── tweet_data_preprocessing.py   # Text cleaning, tokenization, and dataset splitting
├── news_data_collection.py       # Scripts for fetching financial news/data
├── Msc_Project_2792443s.pdf      # Full Research Thesis Report
└── README.md                     # Documentation