Fine-tuned ALBERT model for 3-class sentiment analysis (Negative / Neutral / Positive) on 500k+ tweets, achieving 92.4% accuracy. Includes comprehensive explainability with SHAP, LIME, and Captum (Integrated Gradients) — compared side-by-side with visualizations and top-word analysis. Also evaluated on IMDb reviews for cross-domain robustness.
-
Updated
Dec 8, 2025 - Jupyter Notebook