Dream Content Discovery from Social Media Using AI

Abstract Dreaming is a fundamental but not fully understood part of human experience. Traditional dream content analysis practices, while popular and aided by over 130 unique scales and rating systems, have limitations. Often based on retrospective surveys or lab studies, and sometimes on in-home dream reports collected over some days, they struggle to be applied on a large scale or to show the importance and connections between different dream themes. To overcome these issues, we developed a new, data-driven mixed-method approach for identifying topics in free-form dream reports through natural language processing. We applied this method on 44, 213 dream reports from Reddit's subreddit, where we uncovered 217 topics, grouped into 22 larger themes: the most extensive collection of dream topics to date. We validated our topics by comparing it to the widely-used Hall and van de Castle scale. Going beyond traditional scales, our method can find unique patterns in different dream types (like nightmares or recurring dreams), understand topic importance and connections (like finding a greater predominance of indoor location settings in Reddit dreams than what was in general stipulated by previous work), and observe changes in collective dream experiences over time and around major events (like the COVID-19 pandemic and the recent Russo-Ukrainian war). We envision that the applications of our method will provide valuable insights into the complex nature of dreaming and its interplay with our waking experiences.