Contribution to the Project: Taylor Ray generated documentation on the APIs for YouTube (platform used) and Pinterest (platform unused). She wrote Python scripts to collect data from 7/10 social media platforms (YouTube, Twitter, 4chan, 8kun, Parler, Mastodon, and Flickr). She drew samples from the platforms for both the Scam Study and the general COPE-ID database paper. She developed a multinomial Naive Bayes model for predicting misinformation-related features of COVID-19 posts for the Scam Study portion of the project. She also heavily researched different sentiment analysis techniques to investigate which techniques are best suited for our social media data (VADER Sentiment, multinomial Naive Bayes, Hugging Face’s pipeline, etc.).
Taylor also wrote a technical blog on the VADER Sentiment tool, “Using vaderSentiment to Intuitively Predict the Sentiment of Social Media Posts”, published on the DS3 SSRC lab website. Likewise, she presented “Using Sentiment Analysis Techniques to Discover Emotions Conveyed on Twitter and Reddit” at Mississippi State University’s Spring 2021 Graduate Student Research Symposium and won 1st place in the master’s division of oral presentations for physics, mathematics, and computational sciences/engineering.
In addition, she developed a web application, displayed on the COPE-ID home page, using the Flask web framework that allows users to generate the sentiment polarity (i.e., “positive”, “neutral”, or “negative” nature) and/or misinformation-related features of a single COVID-19 social media post by utilizing the logic of two separate machine learning models. She deployed the web application on Google’s cloud service, Google Cloud Platform. Finally, during this project Taylor worked on her master’s thesis, which was based on the start to finish process of applying a multinomial Naive Bayes model to extract the sentiment polarities of the COVID-19 posts collected for this project.
https://orcid.org/0000-0002-0056-5031