Mo'Mentum

Updated slides including comments/suggestions up to 02/07/2017

Transcript

1. Consulting Proposal September 4, 20XX Mo’Mentum Optimize your mo’mentum for social change Peter Winslow
4. Many resources dedicated to sharing user-generated petitions!
5. Many resources dedicated to sharing user-generated petitions!
6. Victory! Ok, let’s do this! Many resources dedicated to sharing user-generated petitions!
7. Is my petition good enough? Many resources dedicated to sharing user-generated petitions! Not much help for writing them though...
8. Many resources dedicated to sharing user-generated petitions! Not much help for writing them though... Just what I was looking for! Mo’Mentum
9. Many resources dedicated to sharing user-generated petitions! Not much help for writing them though... Just what I was looking for! Mo’Mentum Probability of success
10. Many resources dedicated to sharing user-generated petitions! Not much help for writing them though... Just what I was looking for! Mo’Mentum Probability of success Time scale to reach signature goal
11. Mo’Mentum under the hood: Data Source Change.org sitemap Petition urls
12. Change.org sitemap Petition urls Petition id’s Mo’Mentum under the hood: Data Source
13. Change.org sitemap Petition urls Petition id’s Petition text and metadata Change.org api Mo’Mentum under the hood: Data Source
14. Change.org sitemap Petition urls Petition id’s Petition text and metadata Change.org api Over ~ 40,000 Petitions Mo’Mentum under the hood: Data Source
15. Mo’Mentum under the hood: Feature Engineering Text data Stopwords, Lemmatization
16. Mo’Mentum under the hood: Feature Engineering Sentiment, POS Tagging, word/sentence counts, ... Text data Stopwords, Lemmatization Features
17. Mo’Mentum under the hood: Feature Engineering Sentiment, POS Tagging, word/sentence counts, ... Text data Stopwords, Lemmatization Features Metadata Timestamps, Signature count Success/Failure
18. Mo’Mentum under the hood: Feature Engineering Metadata Timestamps, Signature count Success/Failure Signature Count tfinal - tinitial Targets Sentiment, POS Tagging, word/sentence counts, ... Text data Stopwords, Lemmatization Features
19. Predict success/failure of petition Random Forest Classifier Train-Test-evaluation split with 5-fold CV Model Performance
20. Predict signature accumulation rate Gradient Boosting Regressor: Least Squares loss function Train-Test-evaluation split with 5-fold CV Model Performance
21. About me Peter Winslow The Professional PhD + 1 Postdoc in theoretical High Energy Physics and Cosmology. Specific interests: Origin of matter in the Universe Peter Winslow The New Father! Kiana Winslow, born Nov. 29th 2016
22. Algorithms: Classification Random Forest Classifier (Scikit-Learn) Predict success/failure of petition 18 features after selection Reasons for choosing: • Lots of complication yet resistant to overfitting Challenges: • Class imbalance in the data Validation: Train-Test-evaluation split with 5-fold CV Backup Slides
23. Algorithms: Regression GradientBoostingRegressor (Scikit-Learn) Predict signature accumulation rate 17 features after selection Reasons for choosing: • Many features, highly non-linear, can return predicted “quantiles” Challenges: • The right evaluation metric? Validation: Train-Test-evaluation split with 5-fold CV Backup Slides
24. Insights: Backup Slides New Year’s resolution boom and bust?