Depression, bipolar disorder, post-traumatic stress disorder: these are just some of the mental health conditions which psychoanalysts may be able to track by sifting tweets.
Researchers have already come up with the idea of using the Twitter microblogging network to track the spread of epidemics, especially flu. Now computer scientists at Johns Hopkins University in Baltimore, USA have gone a step further and have set out to identify trends in mental illness. They have developed an algorithm which spots Twitter users who either refer clearly to a mental condition or leave clues to their state of mind in their tweets. In addition to key words associated with the symptoms of mental health problems, the algorithm also takes into account such seemingly innocuous phrases as “I just don’t want to get out of bed.” Having closely examined over eight billion tweets, the Johns Hopkins team have been able to detect some global trends in mental health, ranging from bipolar disorder to depression. One issue springs to mind immediately, however: is the data they collect anonymous? The researchers say it is, stressing that their aim is not to identify individual users who are suffering from a particular condition but rather to learn about global trends, especially from a geographical viewpoint. However, methods already exist to monitor such trends.
Refining existing methods
Psychiatrists traditionally use surveys to get to grips with the main trends as regards clinical depression, stress and post-traumatic stress disorder in populations. The Baltimore team are not planning to replace the techniques already in use but rather to work in tandem with them to improve accuracy. This is pretty much a first, as the majority of researchers running projects of this type are usually looking to supersede traditional methods. At all events, the Twitter-based system will serve to add to the results of the surveys that have been used up to now and will certainly provide faster results. However, the information gathered using this new approach is not at all surprising in itself. It seems perfectly logical that clinical depression is more widespread in areas with high rates of unemployment and that post-traumatic stress disorder is more prevalent on military bases which accommodate veterans from service in Iraq and Afghanistan.
How reliable are the results?
One good reason why the Johns Hopkins team are not seeking to entirely replace existing trend-monitoring methods is that the Twitter-based approach has not proved its reliability. The computer scientists imply this when they point out that collecting data on mental health conditions is far more complex than for tracking viral epidemics. “It's much tougher and more time-consuming to collect this kind of data about mental illnesses because the underlying causes are so complex, and because there is a long-standing stigma that makes even talking about the subject all but taboo,” explains Glen Coppersmith, a senior research scientist at Johns Hopkins who is working on the project. Identifying these conditions through tweets, which are basically public statements, cannot therefore be counted on as 100% reliable and we are still some way away from a Twitter that is “the quiet therapist to whom we reveal much more than we realise ”.