Depression detection using semantic representation based semi-supervised deep learning

Depression detection has become an arduous task in social media due to its complicated association with mental disorders. This work focuses on extracting the depressive features in the social network from the unstructured and structured data through the semantic representation and semi-supervised deep learning model for depression detection (SSDD). The proposed approach primarily performs the hybrid features analysis, unsupervised learning-based depression-influencing features representation, and supervised learning-based depressed user detection processes. Initially, the SSDD investigates the different demographic and content-based features from syntactic and semantic relations. Secondly, adopting the deep autoencoder as the unsupervised learning model leverages the extraction of the depression-indicative features representing the texts with the word embedding. Finally, it determines the depressive texts using the bi-directional long short-term memory (Bi-LSTM) model and facilitates the detection of depressed social users by analysing the profile features, detected depressive tweets, and hybrid knowledge. The experimental results outperform the existing depression detection model.