Predicting political views from text
summary
Claim: basic machine learning and NLP applied to parliament speeches and party manifestos can predict political affiliation with concrete numbers: about 60% for five-way classification and approaching 90% for a binary government vs opposition split, but accuracy collapses on very short sentences and on out-of-domain data. The pipeline uses bag‑of‑words, TF‑IDF, stemming and a simple multinomial logistic regression (scikit‑learn, pandas). A web demo is implemented; it helps exploration but often mislabels and is not yet reliable for automated decisions.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 3:12Problem definition
- 7:51Data sources
- 12:53Initial results
- 17:33Building the demo
- 22:15Model features
- 24:26Sentiment analysis
- 27:14Summary
- 28:55Q&A
- 31:19Q&A
- 34:15Q&A
- 35:48Q&A
- 36:32Q&A
- 37:14Closing remarks
related talks
23mText mining online data with scikit-learn
22mFacebook data and machine learning predicting elections?
38mSentence embeddings for automated factchecking
29mForecasting Australia's 2019 Election with PyMC3
39mSentiment Analysis, its techniques and applications
46mPractical Machine Learning and Rails
34mApplying Machine Learning Techniques to Detect and Understand Fake News
48mLearning Machine Learning