Individual Submission Summary
Share...

Direct link:

Deep Learning the Hidden Victims of Coerced Sterilization with Data Augmentation

Tue, September 28, 2:00 to 3:30pm PDT (2:00 to 3:30pm PDT), TBA

Abstract

Coerced sterilization mainly targets women who are marginalized people in their societies, diagnosed with a mental illness, disability, a history of substance abuse, low-income status, HIV positive. Many scholars and advocacy NGOs recognize that it is a violation of women’s reproductive rights. However, coerced sterilization is a polarized topic. Arguments supporting coerced sterilization include, but are not limited to, reducing the spread of HIV, preventing children born to women suffering from or with a history of drug addiction, and reducing the financial costs on taxpayers. But, there is no systematic empirical research on the topic.



Our study focuses on analyzing this polarized public opinion of compulsory sterilization of women. Relying on online communication platform and social media, we take a supervised machine learning approach to measure opinions. In particular, given the dearth of quality training data, we propose data augmentation techniques to increase the training data in order to improve model performance. We find that data augmentation provides a way to increase the size of the training data and it allows for more accurately measuring the public opinion on controversial topics where data is limited. With the additional training data, we are also able to decrease the costs associated with measuring public opinion while improving model performance.

Author