The goal was to build a high-performing sentiment analysis model that could classify financial news into sentiment categories, supporting market trend forecasting and investment strategy.
I’m passionate about using NLP and deep learning to extract meaning from unstructured data. This project shows how AI can uncover emotional signals in market conversations.
Results
This project involved building a multi-class sentiment model using TensorFlow and Keras, achieving 84% accuracy. I applied tokenization, TF-IDF, and bi-gram extraction to create a robust feature set, then used Keras Tuner to fine-tune hyperparameters.
The result was a production-ready model that can be integrated into real-time dashboards, providing daily sentiment updates from news and social media for financial analysts.
Results
Sentiment data is inherently noisy and imbalanced. Ensuring the model generalized well across domains while avoiding overfitting required continuous tuning, cross-validation, and feature selection.
