Detecting Propaganda in Fake News using Natural Language Processing
summary
A prototype extracts top phrases with RAKE keyword extraction, queries Google News API for matching articles, and measures overlap with spaCy similarity; it then runs event clustering with DBSCAN and flags propaganda using the Path Model of Blame. Tested on WhatsApp forwards and articles (example: a Yes Bank forward during the PMC Bank panic), the MVP catches many simple for‑wards but only on strict thresholds and simple SVO patterns; 50 seeded stories produced ~500,000 reactions illustrating scale, and accuracy falls on noisy or complex sentences.
AI-written summary · about
chapters (21)
- 0:00Intro
- 0:52What is Fake News
- 3:28Statistics
- 6:20Example
- 9:18Identifying Fake News
- 12:26Basic NLP Libraries
- 13:04Finding credible sources for keywords
- 14:32Finding similar keywords
- 14:54Path Model of Blame
- 17:56Location Labelling
- 19:05Extracting Events
- 20:10Entity Extraction
- 22:00Verb Patterns
- 22:36Tags
- 23:06Sample Data
- 24:57The Aim
- 26:25Questions
- 27:50Finding the Source
- 28:29Measuring Value
- 30:02Fake News
- 34:14WhatsApp Texts
related talks
34mApplying Machine Learning Techniques to Detect and Understand Fake News- 34mApplying Machine Learning Techniques to Detect and Understand Fake News
05mEvolving Landscape of Disinformation - Lessons Learned So Far!
31mProvenance analysis against fake news
45mDetecting Plagiarism in News
22mFighting against Marketplace Scams using Community-driven AI
28mNLP Solutions for infodemic management
19mDetection & Classification of Fake news using Convolutional Neural Nets