Sentiment analysis of real-world migraine tweets for population research

Abstract

Background: Migraine is a highly prevalent disorder that is typically episodic in nature. Social network data reflecting personal commentary on everyday life patterns, including those interrupted by migraine, represent a unique window into the real-life experience of those willing to share them. The experience of a migraine attack might be captured by twitter text data, and this information might be used to complement our current knowledge of activity in the general population and even lead to enhanced prediction. Objective: To characterize tweets reporting migraine activity and to explore their social-behavior features as foundation for further investigations. Methods: A longitudinal cohort study utilizing 1 month of Twitter data from November to December 2014 was conducted. Tweets containing the word “migraine” were extracted, preprocessed, and managed using natural language processing (NLP) techniques. User behavior profiles including tweeting frequencies, high-frequency words, and sentimental presentations were reported and analyzed. Results: During the observation period, 98,622 tweets were captured from 77,335 different users. The overall sentiment of tweets was slightly negative for expressive tweets but neutral for informative tweets. Among posted negative expressive tweets, we found a strong tendency that high-frequent expressions were those with the extreme sentiment, and profanity was common. Conclusions: Twitter users with migraine showed distinct sentimental patterns while suffering from disease onsets exemplified by posting tweets with extreme negative sentiments.