Machine Learning for Depression Screening in Online Communities

Abstract

Social media writings have been explored over the last years, in the context of mental health, as a potential source of information for extending the so-called digital phenotyping of a person. In this paper we present a computational approach for the classification of depressed social media users. We conducted a cross evaluation study based on two public datasets, collected from the same social network, in order to understand the impact of transfer learning when the data source is virtually the same. We hope that the results presented here challenge the research community to address more often the issues of reproducibility and interoperability, two key concepts in the era of computational Big Data.