This paper investigates the biases present in Bengali sentiment analysis tools, particularly those influenced by the impacts of colonialism. The authors conducted an audit of all such tools available on the Python package index (PyPI) and GitHub, focusing on identity categories most affected by colonialism – gender, religion, and nationality. The study found that these tools exhibit bias between different identity categories and respond differently to various ways of identity expression. The findings highlight the need for understanding and addressing the colonial influences present in sociotechnical systems like sentiment analysis tools.
Publication date: 22 Jan 2024
Project Page: https://doi.org/10.1145/nnnnnnn.nnnnnnn
Paper: https://arxiv.org/pdf/2401.10535
