This project predicts Myers-Briggs Personality Type (MBTI) dimensions based on user-generated text using Natural Language Processing (NLP) and deep learning techniques. By leveraging pre-trained BERT models, the system classifies personality dimensions into binary categories (e.g., Introversion vs. Extraversion) with improved accuracy and robustness.
The dataset is sourced from Kaggle's Myers-Briggs Personality Type Dataset. It includes approximately 8,600 samples of user social media posts, each labeled with one of 16 MBTI types.
Key steps:
- Analyzed and addressed class imbalance issues.
- Expanded data from ~3,000 entries to ~57,500 samples using text chunking.
- Binary Classification: Simplified MBTI classification into four binary dimensions:
- Introversion (I) vs. Extraversion (E)
- Intuition (N) vs. Sensing (S)
- Thinking (T) vs. Feeling (F)
- Judging (J) vs. Perceiving (P)
- Fine-Tuned BERT Model: Customized the pre-trained
bert-base-uncasedmodel for MBTI personality prediction. - Text Chunking: Split long text posts into 128-token chunks, enhancing data utilization and improving accuracy by 2-3%.
- Model Optimization:
- Early stopping with validation monitoring to prevent overfitting.
- Majority voting to aggregate chunk predictions into a single output.
- Baseline Comparison: Benchmarked performance against logistic regression using BERT embeddings.
- Fine-tuned BERT achieved higher precision, recall, and F1 scores than baseline models.
- Significant improvements in predicting dimensions such as Introversion vs. Extraversion.
- Identified challenges in predicting certain dimensions (e.g., Sensing vs. Intuition) due to dataset imbalance.
- Programming: Python
- Deep Learning Framework: PyTorch
- Models: Pre-trained BERT (
bert-base-uncased) - Tools: Scikit-learn, Hugging Face Transformers, NumPy, Pandas
- Visualization: Matplotlib, Seaborn
- Clone this repository:
git clone https://github.com/yourusername/mbti-prediction.git