Demographics in Social Media Data for Public Health Research: Does it matter?

Social media data provides propitious opportunities for public health research. However, studies suggest that disparities may exist in the representation of certain populations (e.g., people of lower socioeconomic status). To quantify and address these disparities in population representation, we need demographic information, which is usually missing from most social media platforms. Here, we propose an ensemble approach for inferring demographics from social media data. Several methods have been proposed for inferring demographic attributes such as, age, gender and race/ethnicity. However, most of these methods require large volumes of data, which makes their application to large scale studies challenging. We develop a scalable approach that relies only on user names to predict gender. We develop three separate classifiers trained on data containing the gender labels of 7,953 Twitter users from Kaggle.com. Next, we combine predictions from the individual classifiers using a stacked generalization technique and apply the ensemble classifier to a dataset of 36,085 geotagged foodborne illness related tweets from the United States. Our ensemble approach achieves an accuracy, precision, recall, and F1 score of 0.828, 0.851, 0.852 and 0.837, respectively, higher than the individual machine learning approaches. The ensemble classifier also covers any user with an alphanumeric name, while the data matching approach, which achieves an accuracy of 0.917, only covers 67% of users. Application of our method to reports of foodborne illness in the United States highlights disparities in tweeting by gender and shows that counties with a high volume of foodborne-illness related tweets are heavily overrepresented by female Twitter users.