Towards quantifying parliamentary discourse (2019) - updated repository. Older versions can be found here.
- code to scrape 9 years of Czech parliamentary discussion, consisting of 384,430 speeches. Scraped from: http://www.psp.cz/eknih/
- tidying, structuring, stop word removal and adding metadata about MPs
- tokenization and lemmatization using udpipe
- lda using vowpal wabbit
- measuring topic usage over time using Kullback–Leibler divergence.
- novelty: how different (surprising) a speech was from the last w ones
- resonance: how lasting (impactful) a speech was over the next w ones
- modelling the relationship between a speech's novelty and resonance using linear mixed-effect models.
- a few plots
Paper introducing the novelty, transience and resonance measures: Barron, Huang, Spang & DeDeo (2018).
Many thanks to Malte Lau Petersen!