Face Mask Sentiment Analysis

Data Collection

The team utilized Twitter as a primary data source to analyze public sentiment on face masks during the COVID-19 pandemic. Jonathan Lai explains that they gathered JSON representations of tweets from January to August, focusing on user attributes like location and biography, while cleaning the data by removing URLs, hashtags, and emojis 1. Jiebo Luo highlights the advantages of using social media over traditional surveys, noting that Twitter's text-based format and large user base provide a more comprehensive and real-time analysis of public opinion 2.

We decided to use social media where we can get lots of data, we can get millions of users, and we can get this data on an ongoing basis to perform real-time analysis.

--- Jiebo Luo

This approach allows for a broader understanding of public sentiment compared to limited-scale surveys.

   

Demographic Analysis

To infer demographic data from Twitter profiles, the team employed machine learning techniques and pretrained classifiers. Neil Yeung describes using a model called M3, which analyzes profile images and textual inputs to determine age, gender, and other attributes 3. Despite the lack of ground truth data, Jiebo Luo asserts that their methods achieve reasonable accuracy, particularly in gender classification, which exceeds 90% accuracy 4.

Even if there is sort of fluctuations in accuracy, as long as it's within an acceptable degree, and you basically have a lot of data like a ton of data.

--- Neil Yeung

These techniques enable the researchers to make substantiated claims about demographics, despite inherent limitations.