This project worked with two datasets from Kaggle containing tweets regarding the conflict between Russia and the Ukraine. The objective was to assess the accuracy of unsupervised sentiment analysis techniques on these datasets. Various ML models were studied but the highest performance was earned by a Linear SVC. All of the models were trained on a dataset that contained Tweets with known sentiment (all containing patriotic hashtags for the Ukraine). The models were then applied to a wider scoping dataset with unknown sentiment.
Various transformations and visualizations were created during the proces. See visualizations
for all the graphs that were implemented using D3.js
. All the code responsible for the analysis can be found in twitter_sentiment_analysis
.
Peter Gish University of San Francisco - CS Department Spring 2022