JMIR Publications

ABSTRACT

Background:

Many people suffer from insomnia, which is a sleep disorder characterized by difficulty in falling and staying asleep during nighttime. Social media platforms enable users to share their thoughts, opinions, activities, and preferences with their friends, family, and acquaintances. Many users share their Insomnia issues on these platforms as well these social media content can be used to diagnose different mental health problems. However, a few studies predict insomnia from Twitter, but we find still missing research where no researcher conducts strong semantic analysis (i.e., word embedding) in their word use pattern of tweets. We also find gap that no studies show correlation between users’ insomnia and their Big5 personality traits from social media interaction.

Objective:

The purpose of this study is to build an insomnia prediction model from users’ psycholinguistic, i.e., word use, pattern and their Big5 personality traits derived from tweets.

Methods:

In this paper, we exploit both psycholinguistic and personality traits derived from tweets to identify insomnia patients. As a first step of the process, we build users’ psycholinguistic profile from word choices and semantic relationship among the words of 1,755 users’ tweets. Then, we find the relationship between users’ personality traits, and insomnia. Finally, we build a two-step double weighted ensemble classification model to predict insomnia from both psycholinguistic and personality traits derived from user tweets.

Results:

Our classification model shows strong prediction potential (75%) to predict insomnia. Insomniac people are generally ill tempered, and irritable. They feel more stressed, angry, sad, and mentally exhausted. They tend to use negate category of words, i.e., no, not, never. These people frequently use swear category of words, i.e., damn, piss, fuck, etc. with strong temperament. They also use anxious (i.e., worried, fearful, nervous, etc.) and sad (crying, grief, sad, etc.) categories of words in their tweets. We also find that the users with high Neuroticism score likely to have strong correlation with insomnia. They tend to undergo through depression, social introversion, repression, and intolerance. We also observe that users with high conscientiousness scores have strong correlation with insomnia pattern. In our study, we find negative correlation between Extraversion Big5 trait and insomnia.

Conclusions:

In this study, we have investigated users’ psycholinguistic and personality association with insomnia. We have also found out semantic relationship among words of users’ tweets by using word embedding technique. Then, we have built a rigorous ensemble model by using the three different models. Our ensemble classifier have shown strong prediction potential (AUC-78%). We have built the classifier by using a novel 2-step weighted ensemble technique that outperforms the independent classifiers. We plan to improve our classifier by integrating more data on social network such as friend list, time of the tweets, gender, work place, time spent on activities, etc. We also plan to analyze tweets with different languages.