Full paper coming soon.
Sentiment Analysis (SA) is a field of natural language processing (NLP) that uses machine learning to analyze text for opinions, sentiments, attitudes, and emotions. Various applications are now using SA, from employee surveillance to babysitter vetting to flagging “risky” texts on your child’s phone to identifying suicidal behaviour on social media platforms.
This ongoing CRAiEDL project examines some of the many unanswered ethical questions surrounding the various practices that contribute to the creation of SA models, with a specific focus on their design and policy implications. Some of the questions we’re asking include: How are annotation decisions made and by who? How justified are claims about a model’s objectivity and generalizability? What are some policy implications for applying SA models responsibly in various decision-making contexts such as healthcare, employment, or parenting? How can we ensure SA models are developed and deployed responsibly?
In answering these questions, we expose a need for, and aim to frame a conversation related to, policy development, best practices, and self-regulation in the field of SA.
Keep posted for our full paper, which we plan to release in early 2020.