Paper Title
Depression Detection from Social Media Tweets using Graph Convolution Networks

Abstract
Just like physical health, mental health plays an essential role in an individual's life. Being mentally healthy helps in being productive, having better relationships, and having a better self-image, but if not taken care of may lead to mental illnesses. Over time social media platforms have become a fundamental part of day-to-day lives, indicating a close relationship with users. As a result, users reflect their personal lives on these platforms, opening a new research area in recent times. In this study, we intend to analyse tweet data using a Graph Convolution Network (Text GCN), which could provide an initial diagnosis of mental health issues with depression as our primary focus. Numerous studies have employed LSTM and Bi-LSTM models, but only limited ones have explored Graph Convolution Networks (GCN) to perform this task. To train a Text Graph Convolutional Network (Text GCN) for a corpus, we first create a single text graph for the corpus using word co-occurrence and document word relations. The Text GCN learns the embeddings for words and documents under the supervision of the recognised class labels for documents after being trained with a one-hot representation for words and documents. Predictive word and document embeddings are learned by Text GCN. This study has been able to produce 75% accuracy. The accuracy could not be improved due to the limitation of word sequence neglection in GCN. Keyword - Depression, Text GCN, one-hot encoding, Sentiment analysis, NLP, Deep Learning, Pre- processing.