Predicting Mental Health Problems with Automatic Identification of Metaphors

Abstract

Mental health problems are prevalent and an important issue in medicine. However, clinical diagnosis of mental health problems is costly, time-consuming, and often significantly delayed, which highlights the need for novel methods to identify them. Previous psycholinguistic and psychiatry research has suggested that the use of metaphors in texts is linked to the mental health status of the authors. In this paper, we propose a method for automatically detecting metaphors in texts to predict various mental health problems, specifically anxiety, depression, inferiority, sensitivity, social phobias, and obsession. We perform experiments on a composition dataset collected from second-language students and on the eRisk2017 dataset collected from Social Media. The experimental results show that our approach can help predict mental health problems in authors of written texts, and our algorithm performs better than other state-of-the-art methods. In addition, we report that the use of metaphors even in nonnative languages can be indicative of various mental health problems.