This week, we will use the techniques for recommender systems in an unexpected way to help us model topics found on Twitter. In this homework you will practice extracting topics from tweets using matrix factorization. This method assumes every tweet is a combination of several topics weighted by their prevailance in the text. This approach in fact finds a low-dimensional representation of the tweets (through the topic weights).
For this assignment, we will be working with tweets about the pandemic from over a year ago when the pandemic recently entered our lives. The dataset is obtained from Kaggle and the preprocessing we have done followed the steps here. For computational speed we will analyze a dataset from one day: April 30, 2020. We encourage you to explore this dataset further and see how topics change over time.
Any cell with #TODO is code from myself, Cecilia Barnes. All other code from this assignment belongs to the copyright below.
Copyright ©2021 Valentina Staneva Hunter Schafer. All rights reserved. Permission is hereby granted to students registered for University of Washington CSE/STAT 416 for use solely during Spring Quarter 2021 for purposes of the course. No other use, copying, distribution, or modification is permitted without prior written consent. Copyrights for third-party components of this work must be honored. Instructors interested in reusing these course materials should contact the author.