news classification nlp

Overview

Built an NLP classifier that categorizes news articles into 31 distinct categories using logistic regression with CountVectorizer.

Data Processing

- Dataset: Sourced from Kaggle (news headlines with categories)

- Preprocessing Pipeline:

- Text cleaning (special characters, lowercase)

- Stopword removal (English)

- Lemmatization (WordNet)

- Label encoding (31 categories)

Model Architecture

- Vectorization: CountVectorizer (unigrams)

- Classifier: Logistic Regression

- L2 regularization (C=5.0)

- lbfgs solver

- 5000 max iterations

Performance

Achieved classification through:

- Stratified train-test split (80-20)

- Comprehensive text preprocessing

- Hyperparameter tuning

Deployment

Built a Streamlit web app featuring:

- Clean, responsive UI with custom CSS

- Real-time category prediction

- Error handling and loading states

- Category mapping system

Technical Stack

- Python (Pandas, NLTK, Scikit-learn)

- Streamlit for web interface

- Joblib for model persistence

The project demonstrates end-to-end NLP pipeline development from preprocessing to deployment.

This version is:

1. Professional and technical

2. Concise (easy to read on LinkedIn)

3. Highlights key achievements

4. Structured for readability

5. Focused on measurable outcomes