Panderer's toolBox

Transcript

1. Reaching out to potential candidate evangelists: How do we find them? What should we talk to them about? MESSAGE TAILORING WITH PANDERER’S BOX TOOL MARY THOMAS
2. ALGORITHM Sentiment Coded by Candidate @Mentions Sentiment Coded by Candidate Names #marcorubio giving anuther high-school shakespearian soliloquy. #amnesty scammer #gopdebate listen little boy #TrumpTrain
3. ALGORITHM Sentiment Coded by Candidate @Mentions Sentiment Coded by Candidate Names #StandWithRand #FeelTheBern #TrumpTrain Tweepy Twokenizer, POS tagger NLTK WordNet Lemmatizer 10x Cross- validation of classification
4. ALGORITHM Sentiment Coded by Candidate @Mentions Sentiment Coded by Candidate Names #StandWithRand #FeelTheBern #TrumpTrain Tweepy Twokenizer, POS tagger NLTK WordNet Lemmatizer 10x Cross- validation of classification
5. ALGORITHM Sentiment Coded by Candidate @Mentions Sentiment Coded by Candidate Names #StandWithRand #FeelTheBern #TrumpTrain Tweepy Twokenizer, POS tagger NLTK WordNet Lemmatizer 10x Cross- validation of classification
6. ALGORITHM Sentiment Coded by Candidate @Mentions Sentiment Coded by Candidate Names #StandWithRand #FeelTheBern #TrumpTrain Tweepy Twokenizer, POS tagger NLTK WordNet Lemmatizer 10x Cross- validation of classification
7. HUCKABEE: FAIRTAX + CHRISTIANITY
8. HUCKABEE: FAIRTAX + CHRISTIANITY protect, fairtax, tax
9. THE FATE OF HUCKABEE SUPPORTERS -1 -0.5 0 0.5 1 1.5 2 Trump Cruz Rubio Carson Bush Christie Kasich Fiorina Standardized acquired followers (relative to polls) Candidate
10. THE FATE OF HUCKABEE SUPPORTERS -1 -0.5 0 0.5 1 1.5 2 Trump Cruz Rubio Carson Bush Christie Kasich Fiorina -1 -0.5 0 0.5 1 1.5 2 Trump Cruz Rubio Carson Bush Christie Kasich Fiorina Standardized acquired followers (relative to polls) Standardized message representation on “god” topic Candidate Candidate
11. Ph.D. Cognitive Psychology, University of Michigan Research Scientist, Integrated Brain Imaging Center, University of Washington MARY THOMAS
13. LDA VISUALIZATION
14. INCORPORATING OFF-MESSAGE INFORMATION IS IMPORTANT. 0 20 40 60 80 100 Accuracy Weighted F1 Cross-Validation of Classification (13 candidates) Candidate Messaging Topics User Tweet-based Topics