Detection and Analysis of Mental Health Illness using Social Media

Abstract

realizes his or her own potential, can cope with the normal stresses of life, can work productively and fruitfully, and is able to make a contribution to her or his community”. Mental disorders not only affect the mental attributes of individuals like management of emotions, ability to concentrate and interaction with others but it also affects physical health of individuals. The contributing factors listed by WHO that are considered as determinants of mental disorders are genetics, stress, prenatal infections, exposure to environmental hazards, standards of living, working conditions, and community support. Health systems are yet not able to address the burden of mental disorders. Most of the people with disorders live their whole lives with no diagnosis or correct treatment rather it leads to increasing rate of suicide. More than 90% of people who commit suicide have a pre-existing diagnosis of depression. To address the issue many researchers have worked in automatic detection of mental disorders to help practitioners diagnose and carry out correct treatment. This cannot replace mental health professional for obvious reasons. Recently social media has been a widely used network that connects people around the world. Not only this but people sharing their life events, thoughts through posts, status updates all gather up as a big data resource. This resource is helpful in conducting various researches, analyses including big data and machine learning. In this study, we analyzed six mental health issues using Reddit’s data. The data obtained summarizes; Depression, Anxiety, Bipolar, Bipolar Disorder, Schizophrenia, Autism and Mental Health which is a general class which discusses mental health. The data gathered included text of the users’ reddit post and title of the respective post. Experimentation is done using various deep learning and NLP techniques applied for classification. The first phase of experimentation included data preprocessing and feature extraction using GloVe embedding. The second phase included deep learning techniques such as Convolutional Neural Network, Long-short term memory network, Gated Recurrent Unit, Bi- Long-short term memory network and Bi- Gated Recurrent Unit. In addition to these traditional techniques, pre-trained BERT model and RoBERTa model have been applied. Finally a hybrid framework is presented using hierarchical classification and pre-trained RoBERTa fine tuned on the respective mental health data. The last phase compares results of the baseline deep learning models with the presented framework. The results show that the average accuracy of the hierarchical classification with two level hierarchy gives 84% of accuracy on test data. Moreover the results are compared to the pre-trained RoBERTa which gives 82% of accuracy on test data.