They say that breaking up is hard to do. Now, data scientists know that it’s true. Neil Sedaka songs aside, they know it’s true because, in 2013, with the help of public data sourced from Twitter, they were able to track and listen-in on conversations between 661 couples who were in the process of ending their relationships.
Researchers Venkata Rama Kiran Garimella, Ingmar Weber, and Sonya Dal Cin published the results of their study in this paper: From “I Love You Babe” to “Leave me alone” - Romantic Breakups on Twitter.
Before looking at the findings of the study, there are a few interesting details to consider:
Breakups often resulted in a batch unfollowing: an average 15-20 users
Maybe not so surprising, but couples who breakup tend to be fresher (i.e., crass, brazen, or rude) when compared to couples that were still together
There were higher levels of post-breakup Twitter communication between couples who had high pre-breakup levels of interaction
67% of the time, women were the rejector in the relationship
After a breakup, rejectees became more self-centered. They tried to ﬁnd stability in religion and spirituality, yet they also tended to publicly curse their luck, life, and fate
The longer the couple was together during the study, the less likely they were to breakup
‘Stonewalling’ (i.e., ignoring the other’s messages) was a sign of an impending breakup (38% of the time)
Right before a breakup, there was a decrease in the fraction of messages sent to the partner, and an increase in the fraction of messages to other users
There was increased use of “depressed” terms after the breakup compared to the use from couples who did not breakup
Rejectees were a lot more likely to use depressed terms compared to rejectors. This was true both before and after the breakup.
The study notes, 85% of people will experience a breakup at some point in their life. Maybe in the future, when digital assistants become commonplace, they will pick up on these warning signs and remind us to send flowers and clean up our act at just the right moment! The researchers suggested creating this idea as a kind of early break-up warning system.
Another lesson to be learned is that when it comes to data science, nothing is sacred. The information locked behind our everyday interactions online might be revealing more about us, and human nature, than we could have ever guessed.
Some of the reasons for using Twitter, cited in the paper include:
Data collection was relatively easy, especially when facilitated by platforms like CrowdFlower
The size of data was impressive, and there was a lot to work with
Since users did not know they were being watched, there was less self-reporting bias
The platforms allowed for timely collection of data around the moment of break-up
Having social context in the form of network information.
The noise of data made it difficult to navigate
The lack of well-deﬁned variables made it a challenge to define the experiment
There were diﬃculties in observing other psychological variables
There was limited power for determining causal links
Privacy concerns had to be carefully considered.
Overall, it’s amazing to see what you can do with a chunk of data mined from a public datasource like Twitte