Fake News Detection and Propagation: An NLP Perspective, Slides of Natural Language Processing (NLP)

Download Fake News Detection and Propagation: An NLP Perspective and more Slides Natural Language Processing (NLP) in PDF only on Docsity!

Fake News

Propaganda

Vijay Kumar (20/11/EC/021), Nikhil Kumar (20/11/EC/052), Abhishek Singh (20/11/EC/037)

Overview

• (^) Introduction
• (^) Importance of Studying Fake News Propagation
• (^) Why spread Fake news?
• (^) The Role of NLP in Fighting Misinformation
• (^) Growing problem in recent years
• (^) NLP Techniques for Detection
• (^) Advanced NLP Models in Fake News Detection
• (^) Stance Detection
• (^) Network Analysis
• (^) ETHICAL CONSIDERATIONS
• (^) Multimodal Fake News Detection
• (^) Case Study – UKRAINE CONFLICT MISINFORMATION
• (^) Future Challenges

Importance of Studying Fake News Propagation Influence on Society:

• (^) Impacts public opinion, elections, and social behavior.
• (^) Example: Spread of misinformation during elections like the 2016 U.S. Presidential Election. Technological Challenge:
• (^) Tests NLP and AI capabilities to identify and combat fake news. Interdisciplinary Nature:
• (^) Combines NLP, psychology, sociology, and network science

The Role of NLP in Fighting

Misinformation

1.Detection: Using linguistic features to identify fake news. 2.Classification: Categorizing types of misinformation (e.g., satire vs. propaganda). 3.Fact-Checking: Automating verification with knowledge bases like Wikipedia. 4.Source Analysis: Evaluating credibility of authors and publishers. 5.Propagation Tracking: Monitoring how fake news spreads across platforms.

Characteristics of

Fake News

• (^) Lack of credible sources, unusual URLs, and excessive ads.
• (^) Example: Fake health articles on non-reputable domains.

Structural

Patterns:

• (^) Use of sensationalism, emotional language, and clickbait headlines.
• (^) Example: "You Won't Believe What Happens Next!"

Linguistic

Features:

• (^) Misrepresentation of scientific data or outdated information.
  • (^) Example: False claims about climate change using misinterpreted data. Content-Based Indicators:

NLP Techniques for Fake News Detection 1.Text Classification

• (^) Support Vector Machines (SVM):
• (^) Separates fake and real news based on features like TF-IDF vectors.
• (^) Example: Differentiating between biased and objective headlines.
• (^) Naive Bayes:
• (^) Classifies fake news using probabilistic models.
• (^) Example: Categorizing political propaganda articles. 2. Deep Learning Approaches
• (^) CNNs : Capture local patterns and hierarchical structures in text.
• (^) RNNs (e.g., LSTM): Handle long-range dependencies in news narratives.
• (^) Transformers (e.g., BERT): Identify fake news using contextual embeddings.
• (^) Example: Detecting fake COVID-19 treatments with BERT models. 3. Ensemble Methods
• (^) Voting Classifier: Combines predictions from multiple models.
• (^) Boosting: Sequentially improves weak learners like XGBoost.

Feature Engineering for Fake News Detection 1.Text-Based Features: ⚬ (^) N-grams: Identifies sensational phrases common in fake news. ⚬ (^) Sentiment Analysis: Detects exaggerated emotional tone. ⚬ (^) Example: Articles with overly negative tones about vaccines. 2.Contextual Features: ⚬ (^) Analyze author behavior and social media propagation. ⚬ (^) Example: Tracking bots spreading the same article repeatedly.

Stance Detection in Fake News Propagation

• (^) Stance Detection: Determines the attitude towards a claim (support, oppose, or neutral).
• (^) Techniques:
• (^) Feature-Based: Uses n-grams and sentiment lexicons to detect stance.
• (^) Neural Networks: Attention mechanisms in models like BERT to focus on the target claim.
• (^) Graph-Based Models: Analyze relationships between entities in fake news propagation.

Network Analysis in Fake

News

Graph Representation

• (^) Nodes: Represent entities like users, articles, websites.
• (^) Edges: Represent relationships like shares or citations. Graph-based Techniques
• (^) Centrality Measures: Identifies influential nodes (e.g., key social media users spreading fake news).
• (^) Community Detection: Identifies clusters of users spreading similar content.

Multimodal Fake News Detection

Visual-Semantic Embeddings: Projects text and image features into a shared space to identify inconsistencies.

• (^) Text-Image Fusion:
• (^) Detects inconsistencies between text and images (e.g., mismatched headlines and photos). ViLBERT: Model for analyzing both text and images in news articles.

Case Study – UKRAINE CONFLICT

MISINFORMATION

Challenge: Widespread misinformation about the conflict. Approach: 1.Network Analysis: ⚬ (^) Used graph neural networks to track the origin of fake narratives. 2.Temporal Tracking: ⚬ (^) Monitored the emergence of fake stories like "Bioweapon Labs in Ukraine." 3.Multilingual NLP: ⚬ (^) Models like XLM-R analyzed misinformation in Russian, Ukrainian, and English. ⚬ (^) Outcome: Improved real-time fact-checking for war-related content.

THANK YOU