Predicting Twitch.tv Donations using Sentiment Analysis

Authors

Alexander J. Gilbert, Southern Methodist UniversityFollow
Jason Herbaugh, Southern Methodist UniversityFollow
Feby Cheruvathoor, Southern Methodist UniversityFollow
Ben WilliamsFollow
Alex TozzoFollow

Abstract

Twitch.tv streamers have a rare opportunity to receive immediate feedback from their audience through a real-time chat log that is rife with sentiment information. Tools that can help a streamer understand how they need to influence their audience can be useful in increasing the donations and subscriptions they earn. Although millions around the world stream on Twitch, only a minuscule fraction of these streamers earn a living streaming alone. This paper aimed to provide muchneeded guidance to enable more streamers to succeed. We used stream logs, known as VODs (video on demand), which can be easily accessed through Twitch’s API or web interface, and parsed these logs for chat and donation data. After normalizing the data, we performed sentiment analysis using a combination of VADER, TextBlob, and Flair algorithms. We found that chat sentiment is a useful indicator for predicting the occurrence of donations. The results have set the foundation future researchers and developers can use to create tools and further our collective understanding of stream viewer sentiment.

Recommended Citation

Gilbert, Alexander J.; Herbaugh, Jason; Cheruvathoor, Feby; Williams, Ben; and Tozzo, Alex (2022) "Predicting Twitch.tv Donations using Sentiment Analysis," SMU Data Science Review: Vol. 6: No. 2, Article 6.
Available at: https://scholar.smu.edu/datasciencereview/vol6/iss2/6

Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License