Gustav Gyrst - Data Scientist & Researcher

My areas of interest lies within anything that can make our planet a better place. I believe that part of the recipe for such is good music, clean data, and some helpful algorithms. I am currently working on analyzing public opinion on climate change, and how politicians talk about climate change on social media. #ClimateChange #NLP #DataScience #Analytics #SocialScience #Networks #Music.

Recent Work

How do US Politicians Advertise Climate Change on Meta Platforms?

In this study we collect approximately ~600,000 Meta ads from 520 current congress members and analyze ~19,000 climate change related ads to better understand the socio-demographic traits and temporal dynamics patterns in terms of spending and impressions. We find that Democrats are the only really advertising climate change, however, republicans have better ad performance in terms of impressions per dollar spent. It is only 10% of the politicians that drive the debate of climate change.

What Does it Take to Make it Onto One of the Popular Curated Playlists?

In this project we train a classifier to determine a songs likelihood of going onto one of the top Spotify playlist (E.g., New Music Friday, Mood Booster, Beast Mode etc.) based on the musical attributes of the given song. Our support vector classifier (SVC) scores an accuracy of 82.57% for Mood Playlists and 61.66% for determining songs featuring on the hit music playlist New Music Friday.

Matrix Multiplication: A Comparative Study on Algorithm Performances

This is a comparative study to measure performance differences of matrix multiplication algorithms. We work with square matrices with side lengths defined as n = 2 x with x > 0 and try out recursive and non-recursive algorithms. All algorithms is implemented in Python 3.

Detecting Climate Change Quotes in Danish

In this study we fine-tune a BERT-based sequence classifier to identify climate quotes in Danish. KlimaBERT is trained on quotes from discussion in parliament (2012 to 2020). KlimaBERT has a F1 score of 0.97 and is publicly available on Huggingface for use.

Let's Get In Touch

Feel free to reach out by email and contact me.

  • Address

    Rued Langgaards Vej 7
    Copenhagen 2300
    Denmark
  • Phone

    +45 31 76 90 30
  • Email

    gugy@itu.dk