Financial Sentiment Analysis Dashboard

This project provides a comprehensive financial sentiment analysis tool using natural language processing and machine learning techniques. It includes a web-based dashboard for real-time analysis of financial texts and reports.

Features

• Upload PDF files or paste text for analysis
• Sentiment prediction (Bullish, Neutral, Bearish)
• Sentiment scoring
• SHAP (SHapley Additive exPlanations) analysis for model interpretability
• Attention visualization for understanding model focus
• Interactive visualizations

Installation

1. Clone the repository: git clone https://github.com/ademicho123/financial_sentiment_analysis.git cd financial-sentiment-analysis
2. Install the required packages: pip install -r requirements.txt

Usage

Run the Streamlit dashboard: streamlit run dashboard.py 1.

Open your web browser and navigate to the provided local URL (usually http://localhost:8501) 1.

Upload a PDF file or paste text into the input area 1.

Click “Analyze Text” to see the results

Project Structure

• dashboard.py: Main Streamlit application for the web interface
• financial_sentiment_analysis.py: Core sentiment analysis functionality
• sentiment_predictor.py: Sentiment prediction model and utilities
• shap_analysis.py: SHAP value generation for model interpretability
• requirements.txt: List of required Python packages

Model

The project evaluates multiple machine learning models for sentiment analysis:

• TinyBERT (huawei-noah/TinyBERT_General_4L_312D)
• AdaBoost
• Support Vector Machine (SVM)
• Random Forest
• Naive Bayes
• Logistic Regression
• Neural Network

TinyBERT consistently outperformed other models and was chosen for the final implementation in the Streamlit dashboard.

Data Preprocessing

• Text cleaning (removing punctuation, numbers, emojis)
• Tokenization
• Lemmatization
• Stop word removal

Training

The models are trained using:

• Data augmentation techniques (SynonymAug)
• SMOTE and RandomUnderSampler for handling class imbalance
• For TinyBERT: AdamW optimizer with CosineAnnealingWarmRestarts learning rate scheduler

Evaluation

The model’s performance is evaluated using:

• Accuracy
• Precision, Recall, and F1-score for each class
• Confusion matrix
• Cross-validation scores

SHAP Analysis

• SHAP (SHapley Additive exPlanations) values are used to interpret the model’s predictions, showing how each feature contributes to the output.
• Attention visualization helps users understand which parts of the input text the model focuses on when making predictions.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.