Irregular sleep pattern identification and analysis from social media dataset using hybrid deep learning based attention mechanism

Following an irregular bedtime routine and having different amounts of sleep each night might increase a person's risk of obesity, cardiovascular problems, high blood pressure, insulin levels, and other metabolic problems. Similarly, in recent times, social media platforms have gained popularity among users for sharing their interests, thoughts, and opinions. Through social media activities, researchers have been able to mine the text data generated on these platforms to investigate and understand users' behaviors and habits. In this paper, we examine a total of 2,468,697 tweets to identify users' irregular sleeping patterns (ISP) using psycholinguistic and contextual features from their tweets. We conduct a linguistic analysis to understand the factors influencing users' psychological behavior and word use patterns, and find a correlation with their irregular sleeping patterns. We observe that users who have irregular sleeping patterns use anger, anxiety, death, and future categories of words in their tweets largely. In contrast, users with irregular sleeping patterns tend to use positive emotions, family, and other categories of words in their tweets. Building upon our findings, we develop a hybrid prediction model that predicts users' irregular sleeping patterns from psycholinguistic features with an accuracy of 91%. We examine the application of social media data for the early identification of irregular sleep patterns and their related mental and psychological concerns while investigating design prospects for future health technologies to enhance the monitoring and support of healthy sleep behavior. • Create a new X (Twitter) dataset by filtering users sleeping habit-related keywords. • Investigate the relationship between users' ISP and word usage on social media. • Gain insights into users' psychological behavior when affected by ISP. • Build a hybrid attention-based deep learning model to predict a user's ISP. • Utilize psycholinguistic features and word embedding vectors derived from user tweets.