FTDD: Representation Learning using Fourier Transformations for Depression Detection in Social Media Content

Depression is a prevalent and substantial mental health challenge worldwide, and its presence is frequently noticeable through textual posts on various online social media platforms. In the current internet era, text is highly ambiguous and often encompasses various emotions. Current depression detection solutions lack privacy and robustness. We introduce a "Transformation block" that provides the privacy by transforming the embedding of text into different format while holding the actual semantics of the text to analyse the characteristics of depressive posts. Proposed block allows the deep learning models to identify linguistic markers of depression results in significant improvement of the performance of depression detection. The experimental evaluations and comparative analysis demonstrate the potential of the proposed transformation block in enhancing depression detection, thus offering a promising avenue for addressing mental health concerns through online platforms.