Mining Twitter Data for Behavioral Insights on the U.S. Election

A Q&A with a UCIPT Postdoctoral Researcher

In anticipation of the results of Super Tuesday, were you able to see an increase in the stress of college students at campuses in battleground states?

We have not been tracking the stress level of college students on Twitter at campuses in battleground states. This is something that we could aim to do for future election dates. We are currently building an automated web platform that is able to mine and analyze social media data in order to answer this type of question.

Do college students who lean Republican or Democrat express their stress differently on Twitter? Are there studies or articles you can recommend to your readers on this subject?

We do not know whether students express stress differently based on their political views. However, there are studies that have attempted to measure different types of stress on Twitter (e.g., stress experienced in people’s daily lives vs. post-traumatic stress disorder). This will be an interesting area of research in the future.

If you were to apply predictive science to Donald Trump’s Twitter feed, could you give some examples of tweets that clearly show him stressed out? How would that compare to how some of the other candidates express stress?

It is always interesting to try to use Twitter data to understand a person’s stress level. An interesting web application developed by North Carolina State University allows users to analyze Twitter data (e.g., hashtags) to interpret different types of emotions expressed. Politicians usually have a pretty active Twitter feed, so you can use this application to see what types of emotions are being expressed by the various candidates.

Looking at Twitter Analytics, what are the key metrics for analyzing stress? Is it the number of tweets in a certain period of day? An increase or decrease in impressions? Could you provide graphics of the Twitter Analytics of a typical stressed-out college student?

In a previous study, we found that the sentiment of tweets (positive, negative, neutral) can be associated with a person’s stress level. I think in order to accurately predict someone’s stress level, it is important to use multiple relevant predictor variables. Previous studies have found that more social media activity during the evening is associated with greater levels of negative emotion. Currently, there are many Twitter Analytics programs on the Internet with different UI designs. One of the most interesting analytics apps out there, informally known as a Twitter sentiment analysis tool, provides a visual analysis of Twitter keywords.

How has your research influenced your own personal Twitter habits?

I think it has made me more aware of the overall sentiment of my tweets. However, it is important to recognize that there are many different types of Twitter users. Some use Twitter for personal reasons (e.g., sharing their daily activities), while others use it primarily for business. Of course, this influences the type of content that gets posted. It will be interesting to study what sort of influence the type of Twitter user has on the accuracy of health outcome predictions using social media data.

Leave a comment

Your email address will not be published. Required fields are marked *