Sentiment Analysis for Mental Health Using NLP and Deep Learning

Table of Contents

1. Project Overview
2. Installation
3. Usage
4. Data Preprocessing
5. Exploratory Data Analysis (EDA)
6. Model Architecture
7. Training and Evaluation
8. Results Visualization
9. Prediction Function
10. Future Improvements
11. Contributing
12. License

Project Overview

This project aims to predict mental health status based on textual statements using Natural Language Processing (NLP) techniques and a Convolutional Neural Network (CNN) model. The project includes data preprocessing, exploratory data analysis, model training, evaluation, and a prediction function for new statements.

Installation

To run this project, you need to have Python installed on your system. Clone the repository and install the required packages:

git clone [https://github.com/kknani24/Sentiment-Analysis-for-Mental-Health-Using-NLP-and-Deep-Learning.git]
cd sentiment-analysis.ipynb
pip install -r requirements.txt

The requirements.txt file should include:

pandas
plotly
nltk
scikit-learn
textblob
numpy
wordcloud
matplotlib
tensorflow

Usage

To run the main script:

python sentiment-analysis.ipynb

Data Preprocessing

The data preprocessing steps include:

1. Loading the data from 'Combined Data.csv'
2. Handling missing values
3. Text cleaning:
   • Lowercasing
   • Removing text in square brackets
   • Removing links and HTML tags
   • Removing punctuation and newlines
   • Removing words containing numbers
4. Tokenization and stopword removal
5. Data augmentation using translation

Exploratory Data Analysis (EDA)

The EDA phase includes:

1. Displaying basic dataset information
2. Visualizing the distribution of mental health status
3. Analyzing text length distribution
4. Creating a word cloud of cleaned statements
5. Visualizing the proportion of each status category

Model Architecture

The CNN model architecture:

model = Sequential([
    Embedding(input_dim=10000, output_dim=128),
    Conv1D(filters=128, kernel_size=5, activation='relu'),
    GlobalMaxPooling1D(),
    Dense(128, activation='relu'),
    Dropout(0.5),
    Dense(len(label_map), activation='softmax')
])

Training and Evaluation

The model is trained using:

• Optimizer: Adam
• Loss function: Sparse Categorical Crossentropy
• Metrics: Accuracy
• Epochs: 10
• Validation split: 0.2
• Batch size: 32

Evaluation metrics include:

• Test Accuracy
• Classification Report
• Confusion Matrix

Results Visualization

Results are visualized using Plotly:

• Histogram of mental health status distribution
• Text length distribution
• Confusion matrix heatmap
• Word cloud of cleaned statements
• Pie chart of status category proportions

Prediction Function

A predict_status function is provided to make predictions on new statements:

predicted_status = predict_status(statement_to_predict, tokenizer, model, label_map, reverse_label_map)

Future Improvements

Potential areas for improvement:

1. Fine-tuning hyperparameters
2. Experimenting with different model architectures (e.g., LSTM, Transformer)
3. Incorporating more features (e.g., sentiment analysis scores)
4. Collecting more diverse data to improve model generalization

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.