Music listening, mental health and stress: A computational framework for personalized analysis and recommendation

This study looks at how listening to music relates to feeling stressed among people who talk online about living with mental health conditions such as depression, anxiety, post-traumatic stress, or bipolar disorder. Instead of running a lab experiment, we used public posts from a large group of Twitter users. We compared people who showed signs of these conditions with similar users who did not, so the groups were as alike as possible in other ways. We also followed people over time, because the same person can share many posts and their stress can change from day to day.

We focused on two things: the kind of music people listened to and the “mood” and sound qualities of the songs. For music type, we looked at common genres such as pop, rock, hip hop, and jazz. For sound qualities, we looked at features like how fast a song feels, how cheerful or sad it sounds, and whether it is mostly instrumental.

What we found is that music does not relate to stress in the same way for everyone. The pattern depended on both the listener’s mental health group and the type of music. For example, pop music was linked to higher stress later on for some groups, while other groups did not show the same change. Songs that sounded sad or low in positivity were linked to stress rising after a delay for some listeners, while songs that sounded happier were not linked to the same increase.

Finally, we used these patterns to build a simple recommendation approach that does a better job of putting potentially stress-reducing songs near the top of a list, compared with a basic baseline method. Overall, the findings suggest that music choices and their emotional tone may matter differently depending on someone’s mental health context, which points to the potential for more personalized, data-informed music support for well-being.Ghenai, A., & Noorian, Z. (accepted). Music listening, mental health, and stress: A computational framework for personalized analysis and recommendation. ACM Transactions on Information Systems.