With the worldwide prevalence of obesity doubling over the past 30 years across a wide range of demographic and socioeconomic groups, the obesity epidemic is a major public health challenge today. Excessive food consumption and lack of exercise are two major contributors to obesity. In recent years, these two activities have been integrated into our daily lives together with social media. A growing body of literature delineated how social networks impact people's health-related behaviors and suggested social media has a role in affect people's obesity-related behaviors. This study aims to gain an insightful understanding of which online social factors are impacting users' obesogenic behaviors and explore computational methods to examine those behaviors using social media data.
Our work consists of three overall research aims. In the first aim, a systematic review was conducted to examine online social factors concerning obesity. A total of 1,608 studies that related to obesity and social media were collected from the three most popular electronic databases for the field. After close inspection, 50 studies were further examined and ten types of online social factors were identified within four-level social-ecological model, which was used to explain each factors' potential impact on an individual from varying levels of online social structure to user's connection to the real world. In the second aim, we learned how the local food environment influences state level obesity rate using social media data. Publicly available Yelp and MyFitnessPal data were collected via a novel approach to characterize the local food environment. A statistically significant correlation between the state's food environment and state obesity rate was observed. We further built a computational model to predict the state-level obesity rate using aforementioned data, in which we achieved a Pearson correlation of 0.791 across US states and the District of Columbia. In the third aim, we studied how a major social disrupting event, COVID-19 shutdown, affects users' dietary behavior using social media data. Tweets relating to people's dietary behavior with images from April to June of 2019, 2020, and 2021 were collected. An observational study of behavior patterns was conducted by using image classification models, visualization tools, and text analysis methods. People are found eating more healthier food during complete and partial shutdowns than before Covid-19. Results of this dissertation could help the public health agencies, policymakers, organizations and health researchers to better utilize social media to carry out obesity-related education, obesity surveillance, and develop public health policy to address this challenge.