Facebook data and machine learning predicting elections?
summary
Hand-collected Facebook data over three months from 41 French cities covering 225 candidates feeds 25 variables (normalized per voters) into four ML methods: decision trees, KNN, Naive Bayes and neural networks. Neural networks produce the best predictive accuracy while Naive Bayes performs worst; the total page likes is the single strongest predictor, link likes and number of statuses follow. Model errors compare closely with official pre-election polls on a 44-candidate subset. Limits: one country, one election (March 2020), only Facebook.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:18Research question
- 7:14Data collection
- 12:13Machine learning models
- 14:41Polling vs predictions
- 16:32Key predictors
- 17:58Campaign advice
- 19:27Limitations
- 21:15Conclusion
related talks
54mOn How Machine Learning and Auction Theory Power Facebook Advertising
22mGenerating a Billion Personal News Feeds
34mWhat will a whole nation do? Predicting events in human networks
23mKeynote: Digital Psychometrics and it's Future Effects on Technology
37mPredicting political views from text
48mLearning Machine Learning
25mMachine learning algorithms as tools for student success prediction
01h 03mHow to Steal an Election