A Mixed Deep Learning Based Model to Early Detection of Depression

Abstract

Mental health is considered as one of today’s world’s most prominent plagues. Therefore, our work aims to use the potential of social media platforms to solve one of mental health’s biggest issues, which is depression identification. We propose a new deep learning model that we train on a depression-dedicated dataset in order to detect such mental illness from an individual’s posts. Our main contributions lie in the three following points: (1) We trained our own word embeddings using a depression-dedicated dataset. (2) We combined a Convolutional Neural Networks model with the Message-level Sentiment Analysis model in order to improve the feature extraction process and enhance the model’s performance. (3) We analyzed through different experiments the performance of three deep learning models in order to provide more perspectives and insights for depression researches. A total of four classifier models were deployed with the same dataset. Those implementing CNN-BiLSTM with Attention model attained greater overall Accuracy, Recall, Precision and F1 macro scores of 0.97, 0.95, 0.84 and 0.92 on the final assessment test set, respectively.