"Do we like this, or do we like like this?": Reflections on a Human-Centered Machine Learning Approach to Sentiment Analysis

Sourojit Ghosh, Murtaza Ali, Anna Batra, Cheng Guo, Mohit Jain, Joseph Kang, Julia Kharchenko, Varun Suravajhela, Vincent Zhou, Cecilia R. Aragon

Abstract: Machine Learning is a powerful tool, but it also has a great potential to cause harm if not approached carefully. Designers must be reflexive and aware of their algorithms’ impacts, and one such way of reflection is known as human-centered machine learning. In this paper, we approach a classical problem that has been approached through ML - sentiment analysis - through a Human-Centered Machine Learning lens. Through a case study of trying to differentiate between degrees of positive emotions in reviews of online fanfiction, we offer a set of recommendations for future designers of ML-driven sentiment analysis algorithms.