This article was published in Scientific American’s former blog network and reflects the views of the author, not necessarily those of Scientific American
Depending on your perspective, Twitter can either be a valuable source of breaking news, or a fire hose of miscellaneous, often dubious information. Microsoft researchers are investigating whether the microblogging service could serve another, more scientific function—to spot signs of postpartum depression in new mothers based on changes in how and what they tweet.
The research is in its early stages and in some ways relies heavily on data that’s easy to misinterpret. Yet the experiment is noteworthy both for its objective—to potentially identify and assist young families in distress—and for the idea that social media might be mined for the good of social science. An added benefit could be the development of apps installed on smartphones, tablets and computers that can monitor tweets, flag warning signs and discretely offer assistance to women who otherwise might suffer quietly.
“What’s exciting is that we could identify individuals potentially at risk for having an emotional downturn just by looking at streams of publicly shared data,” in this case Twitter feeds, says Eric Horvitz, managing co-director of the Microsoft Research lab. Horvitz and his colleagues presented the results of their efforts to predict postpartum emotional and behavioral changes via social media this week at an Association for Computing Machinery conference on Human Factors in Computing Systems in Paris.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
This postpartum study is part of a larger effort to use social media as a sensor network for public health, Horvitz says. In the past, Microsoft Research has helped create systems that can predict the likelihood a patient will contract an infection while in the hospital or that a hospital patient being discharged will soon be readmitted. Another project demonstrated the ability to detect previously unknown drug interactions by analyzing anonymized Web-search logs that include tens of millions of queries sent to search engines by millions of users. (pdf)
The researchers are not claiming they can diagnose postpartum depression. However, Microsoft’s team does say it was able to identify 376 Twitter users as new mothers and use machine learning software to predict—with 71 percent accuracy—which of these women would exhibit significant changes in their postpartum use of the social network. “We studied the language the women used as well as how many re-tweets they were involved in, whether they were actively re-sharing different external links to other Web sites or whether they were engaging in one-to-one interaction with folks on Twitter,” says Munmun De Choudhury, a postdoctoral researcher at Microsoft Research. “We also looked at the structure of their Twitter network—how many people they follow, how many people follow them and how much this changed following the birth of a child.”
To find women for their study, the researchers first created an automated process that sifted through thousands of tweets published between June 2011 and April 2012. The program searched phrases and keywords—height, weight and gender, for example—that suggested a woman had recently given birth. The software picked up on tweets describing labor and listing the height and weight of a baby, for instance. After creating this initial pool of candidates, the researchers used a program to help identify the gender of the tweeter, discarding male names as well as those that could be either male or female. Then, to distinguish between tweets made by new mothers and those made by female family and friends, Microsoft Research turned to Amazon’s Mechanical Turk digital labor marketplace, hiring workers to analyze a candidate’s birth-announcement posting in the context of her tweets before and after the birth.
Microsoft Research examined several months of tweets for each Twitter user they identified as a new mother. Not surprisingly, the new mothers didn’t tweet as often (changing lots of diapers will do that), but the researchers also noted in some users a drop in positive expressions and an increase in negatives—words associated with anger, anxiety and sadness. Women that seemed to experience more profound behavior changes—compared with their prenatal selves on Twitter—also tended to use first-person pronouns more often in their tweets. This may be an indication of isolation and an increasing focus on themselves, says De Choudhury.
The most interesting aspect of this project is yet to come, as Microsoft Research is now working with experts in postpartum depression at the University of Washington and elsewhere to see how their predictive modeling holds up in a population of women that includes those who have been professionally diagnosed with postpartum depression, Horvitz says.
Acknowledging the limitations of this study (none of the new mothers were actually contacted to confirm any of the researchers’ assessments, for example), Horvitz says the predictive models he and his team are testing could someday be used to help design new kinds of early warning systems for women at risk of postpartum depression, even though Microsoft itself would not necessarily develop such technology.
“Postpartum depression is believed to be a very underreported condition,” he says, adding that any efforts to help women recognize their situation and encourage them to seek help would be welcome.
Image courtesy of Ambro at FreeDigitalPhotos.net