FAKE-NEWS-CLASSIFICATION-WITH-LLM-TECHNIQUES.MD
NLP + EXPLAINABLE CLASSIFICATION
Fake News Classification with LLM Techniques
A text-classification project focused on fake/real news detection using multiple NLP representations with explainability and class-balance handling.
Overview
This project focuses on fake news detection through NLP-driven text classification. The main value of the project is not just prediction, but also interpretability and careful handling of imbalanced text data.
Problem
News credibility detection is challenging because model quality depends heavily on representation choices, data balance, and interpretability. A useful classification pipeline needs more than a raw score.
Role
I worked on text representation, classification workflows, and explainability-oriented evaluation so the system could be better understood and improved systematically.
Implementation
- Built text-classification pipelines using TF-IDF and Word2Vec-style representations.
- Applied classification workflows to separate fake and real news examples.
- Used SHAP to improve interpretability around feature impact and model behavior.
- Used SMOTE to support better balance handling during training and evaluation.
Outcomes
- Created a stronger fake/real classification workflow grounded in multiple NLP representations.
- Improved interpretability through SHAP-based explanation layers.
- Strengthened class-balance handling and evaluation discipline with SMOTE-based preprocessing.