This project builds a multi-class sentiment classifier to detect the rhetorical tone of political speech segments, classifying them into four categories: Apathetic, Moderate, Strongly Worded, and Extreme.
- Used TF-IDF vectorization and logistic regression to classify segments based on tone
- Tuned hyperparameters and preprocessing steps to achieve the second highest accuracy in my team
- Contributed to data cleaning, model training, and evaluation
- Explored real-time applications in platforms like Twitch to support healthier chat moderation
Model.ipynb– Notebook with preprocessing, model training, and evaluationInfo 159 Annotation Project.pdf– Official annotation rubric used for labeling training data
Python, Scikit-learn, TF-IDF, NLTK, Jupyter Notebook
This project was completed as part of UC Berkeley’s INFO 159 course. It combines NLP classification with real-world annotation criteria to explore shifts in political rhetoric — and potential extensions to streaming or social platforms.