Doctoral student Zachary Stine presented research about machine learning and the legislative evolution of Ukraine during the EIT Colloquium on March 1. Stine and co-author Dr. Nitin Agarwal explored methods on how to quantify change in Ukrainian legislation over time in order to provide a complementary view into the larger political dynamics of the country.
In order for the audience to understand the legislative evolution of Ukraine, Stine provided an overview of the recent political events. Since its independence from the Soviet Union in 1991, Ukraine has gone through several political challenges. The ongoing conflict with Russia became a global focal point when Russia annexed Crimea in 2014. The country has been internally divided into those who support closer ties to Russia and others who prefer the country joining the European Union.
The research began with the obtainment of Ukrainian draft laws spanning a 12-year period from the Ukrainian parliament’s website. This was done with the help of a scraping tool. The researchers then applied probabilistic topic modeling using an algorithm called latent Dirichlet allocation (or LDA). “After running LDA, we represented each draft law as a distribution of topics, which we ordered chronologically.” Stine explained. “We then compared how surprising a given law’s topic distribution is in light of the preceding laws using an information theoretic measure called the Kullback-Leibler divergence. This allowed us to calculate how novel a given draft law is from the previous laws and thus identify periods of heightened and lessened legislative novelty over time.”
The most interesting finding to the researchers is that the parliamentary committees most responsible for injecting legislative novelty into the parliament are the committees which deal with Ukraine’s relationship with the European Union and the committee on foreign affairs. Thus, the primary context of Ukrainian legislative innovation deals with how Ukraine relates to other countries and alliances. “The biggest challenge in doing this study was simply gathering all of the data.” reflected Stine. “I had to write a program to download all the documents, which were Word files, and then had to write a separate program to strip out all of the text from those Word documents and convert them to TXT files for processing, which was a bit of a headache.”