Machine Learning Finds “Fake News” with 88% Accuracy
You can read this in 53 seconds.
- Finding Fake News is a great task for machine learning
- Editor: See the Viral Bot post here
- Used Kaggle’s Fake News data set (13,000 articles)
- Finalized data set = 10,558 articles
- Plan was to go the routine spam detection route
- Used word freq and TFIDF
- Word freq and TFIDF can only get you so far
- Extracted document titles and full text
- Used a Naive Bayes Classifier
- Used Scikit-Learn to tune parameter
- Cross Validated accuracy = 91.7%
- Testing on remaining 5,234 articles (out of sample data), 88.2% accuracy in identifying Fake News
Read the entire source article here.