Using Content Analysis and Machine Learning to Identify COVID-19 Information Relevant to Low-income Households on Social Media

COVID-19 pandemic has caused great distress in the lives of many populations. Low-income households are among the most severely impacted groups in the United States and across the globe. Using social media, this paper aims to identify and organize the information about the impact of the pandemic on low-income households. We use content analysis to derive an annotation protocol and manually annotate a tweet dataset using this protocol. Furthermore, we use machine learning to learn models from the annotated dataset. We also employ a human-in-the-loop data augmentation procedure to improve the model's performance for the underrepresented classes. Our results show that using carefully annotated data, automated machine learning models can be trained and employed to identify information relevant to low income households, potentially in real time.