This project presents a comparative analysis of three sentiment analysis approaches applied to a synthetically generated customer review dataset. The methods evaluated include:
- A custom sentiment classifier built using Natural Language Toolkit (NLTK), TF-IDF vectorization, and a traditional machine learning algorithm
- A rule-based sentiment analyzer using the TextBlob library
- A deep learning-based sentiment classifier based on a pretrained Transformer model (BERT) from the Hugging Face library
The study explores:
- Trade-offs between accuracy, computational efficiency, and interpretability across the three techniques
- Real-world applicability of each approach to customer sentiment analysis
- Visualization of sentiment trends over time using model outputs on synthetic review data