With its millions of users, Twitter has evolved into a huge channel for personal expression, which can be useful in public health issues – especially depression.
Eric Horvitz, of Microsoft Research Redmond, is a pioneer in research on Twitter and depression, and estimates that at some point "intelligent" systems will be able to analyze a user's Twitter feed and warn them if they are at risk of depression, according to a Time report.
"We wondered if we could create tools that would be able to tell when someone is depressed, just from their posts. What do people say to the world in public spaces? Can you imagine tools that would inform about changes in a user's mood, before they even feel them," he says.
Horvitz and a team of researchers have helped develop a method that can predict depression in Twitter users with 70% accuracy by “scanning” their tweets. However, the method still has a lot of room for improvement, as it can miss some clues and fail to identify people who may have the problem. Also, according to Horvitz, there is an issue with errors, as in 10% of cases healthy users were found to be at risk of depression.
For the study, the Microsoft team identified 476 Twitter users, 171 of whom had depression. The researchers then examined their Twitter histories for up to a year before their diagnosis, looking for various signs of depression, analyzing 2.2 million tweets with computational models. By comparing the tweets of depressed users with those of healthy users, they developed a method that can predict cases of depression before they occur. This “model” was then tested on another sample of users, with a success rate of 70%.
Some tweets were "obvious": "I want someone to hug me and be there for me when I'm sad," "Having a job again makes me happy. Less time to be sad and watch sad movies," etc. However, Microsoft researchers also looked at other factors, such as the number of tweets per day, the time users tweeted, the degree of interaction with others, the type of language, etc.
They also looked for keywords that indicate depression ("anxiety," "nausea," "sleep," "nervousness," etc. were frequently used, but there were also other, seemingly more "innocent" words, such as "love," "him," "her," "home," "tolerance," "songs," "movies," etc.). The frequency of tweets and dialogues with other users is also important, as people who are depressed tend to tweet less and interact less with other people, according to Horvitz.
One area where this type of research could be useful is in assessing public reactions to major events. Observing and analyzing Twitter feeds after “traumatic” incidents could allow for further understanding of how users are affected by the news.
“Our view is that Twitter is the largest observational study of human behavior we’ve ever seen, and we’re working hard to exploit it,” Tyler McCormick, of the Center for Statistics and Social Sciences at the University of Washington, tells Time. McCormick’s team is also working on the subject, as is a team at the University of California, San Diego.
Source: naftemporiki.gr
