Vulnerable to misinformation?: Verifi!
summary
Verifi2 visual analytics system exposes concrete signals that separate trustworthy and suspicious news: it ingests over 1 million tweets from 285 labeled Twitter accounts (Oct 25–Jan 25, 2017), extracts linguistic features (sentiment, emotion, moral foundations, subjectivity), social networks (mention/retweet communities), and image embeddings (ResNet18 + Word2vec-style cosine similarity), trains classifiers that flag fairness/loyalty/subjectivity with real accounts and fear/anger/negativity with misinformation, lets users compare named-entity word usage and image framing, and proves useful in expert interviews while recommending a trustworthiness spectrum and user-editable outputs.
AI-written summary · about
chapters (10)
- 0:00Introduction to misinformation
- 1:06Why sources matter in misinformation
- 3:16Introducing Verifi2 system
- 6:05Linguistic features analysis
- 7:29Social network analysis
- 8:51Image comparison across sources
- 11:23Expert feedback and improvements
- 13:09Contributions and future work
- 14:12Call to action and conclusion
- 14:54Q&A session
related talks
11mIdentifying Disinformation Websites Using Infrastructure Features
40mTechnology to counter misinformation/disinformation
32mTricking the Trolls: A Look at the Digital Future
40mHot Topics in Misinformation: Election 2020, COVID 19, and More
29mThe Invisible War to Hijack Minds, Distort the Truth, and Impact Elections
51mMisinformation Is the New Malware
34mApplying Machine Learning Techniques to Detect and Understand Fake News
35mDisinformation or Debate? Finding Signal Amid Political Noise