Understanding Socio-Cultural Factors related to Obesity: Sentiment Analysis on related Tweets

Abstract

ObjectiveWe aim to better understand socio-cultural factors (i.e., attitudes or perceptions of cultural groups) associated with food consumption and weight loss via sentiment analysis on tweets, short messages from Twitter.IntroductionObesity can lead to the death of at least 2.8 million people each year1, yet the rate of obesity around the world has continuously increased over the past 30 years1. Societal changes, including increased food consumption and decreased physical activity, have been determined as two of the main drivers behind the current obesity pandemic2. Examining socio-cultural factors (i.e., attitudes or perceptions of cultural groups)3 associated with food consumption and weight loss can provide important insights to guide effective interventions and a novel surveillance approach to characterize population obesity trends from sociological perspectives. The primary goal of this study is to examine socio-cultural factors associated with food consumption and weight loss by conducting sentiment analysis on related online chatters. The secondary goal is to discuss the potential implications of being exposed to these different chatters in the online environment. Scientific evidence in support of using social media to understand socio-cultural factors and its potential implications can be illustrated in two concise assertions. First, online chatters, including discussions on social media, have been shown to be an effective data source for understanding public interests4,5. Second, prolonged participation in social media has been suggested to have impacts on users6–8.MethodsIn this study, we examined Twitter (www.twitter.com), a highly popular, free-to-use, micro-blogging social media platform that can instantly broadcast short messages to the world. These short messages are called Tweets and we collected weight loss related and food consumption related Tweets using Python library called Tweepy9. We used hashtags from a previous study10, including #weightloss, #diet, #fitness, and #health for collecting weight loss related tweets. Similarly, we used #Food, #FoodPorn, and #Foodie to collect food consumption related tweets. We then used a rule-based model called Vader11, a sentiment analysis tool (i.e., computational process of categorizing sentiment) developed for social media text, to measure tweets’ sentiment. We used the compound score, which is a normalized and weighted composite score that ranges from -1.0 (most negative) to 1.0 (most positive). Lastly, we conducted independent sample t-test to compare the sentiments of two types of tweets.ResultsWe collected 81,535 (from 41,436 unique user ID) weight loss related tweets from August 30th to September 2nd of 2018 and 86,277 (from 36,977 unique user ID) food consumption related Tweets from August 28th to September 2nd of 2018. The mean sentiment score for weight loss related tweets was 0.17 (sample standard deviation: 0.39), whereas the mean sentiment for food consumption related Tweets was more positive, scoring 0.26 with sample standard deviation of 0.34. The independent sample t-test suggests that the sentiment difference between the two types of tweets is statistically significant (t=52.10, p