AI Model, Twitter Data Provide Population-Level View of Physical Activity

Knowing the differences in population-level physical activity can help public health officials design interventions that target the specific needs of communities.

Using machine learning to comb through exercise-related tweets, researchers identified regional and gender differences in exercise types and intensity levels, providing insights into possible interventions that target certain communities, according to the findings of a study published in BMJ Open Sport & Exercise Medicine.

The machine-learning method also allowed researchers to see how different populations feel about different kinds of exercise.

The findings revealed that walking was the most popular physical activity for both men and women across all regions. Men and women also mentioned performing gym-based activities at similar rates, with men mentioning such activities in approximately 4.68% of tweets, compared to 4.13% for women. Among these tweets, CrossFit was the most popular among men’s tweets, showing up in approximately 14.91%. Yoga was more popular among women, being mentioned in 26.66% of tweets.

Women in the West did more intensive exercise than other regions, while the Midwest had the most intensive exercise among men. The South had the biggest gender gap in exercise intensity. The average difference between men and women in the South was 8.51 calories burned per 30 minutes of activity.

“In most cases, lower-income communities tend to lack access to resources that encourage a healthy lifestyle,” said senior study author Elaine Nsoesie, Ph.D., assistant professor of global health at Boston University School of Public Health. “By understanding the differences in how people are exercising across different communities, we can design interventions that target the specific needs of those communities.”

Researchers used artificial intelligence (AI) models to find an analyze more than 1.38 million exercise-related and geography-tagged tweets by more than 481,000 Twitter users in 2,900 U.S. counties. The AI eliminated false positives, like references to The Walking Dead or watching sports. The model used a set of 376 keywords.

The research team then compared the tweets by men and women and from four different regions: The Northeast, the South, the Midwest and the West.

Top exercise terms included “walk,” “dance,” “golf,” “workout,” “run,” and “pool.”

For women in the West, hiking was the second most popular activity, representing approximately 15.18% of tweets. In the Midwest and South, hiking represented only 3.24 to 3.79% of tweets.

Participating in yoga was more prevalent in the Northeast than the South.

When comparing mentions between men and women, the average number of reported calories burned per 30 minutes of exercise for men was approximately 201.27, compared to 191.66 for women.

The counties that reported higher level of physical activity on Twitter had lower physical inactivity prevalence. These correlations were strongest in the Northeast and West.

The findings suggest that digital data, including social media, could provide valuable health behaviors. The study also demonstrates that Twitter is useful for measuring small-area trends in physical activity, which is an important risk factor for non-communicable diseases.

“Monitoring physical activity using social media will allow public health officials to identify changes in health behaviors at small geographical scales across the U.S.,” the study authors wrote. “Findings from this study provide an important step in the right direction.”

Get the best insights in digital health directly to your inbox.

Related

Can mHealth Apps and Text Messages Get Teens to Exercise?

How Digital Health Interventions Can Reduce Sedentary Behavior in Office Workers

Apple Pumps ECG App for Heart Month, in Hot Pursuit of Digital Health Adoption