Depression Detection in Transcripted Text

Description There is a steadily growing body of evidence exploring automated – machine learning-based – methods to aid the diagnosis of mental illnesses such as depression using audio and visual cues. At the same time, evidence from psychology literature suggests that language of depressive individuals contains a higher usage of negative emotion and first-person words, most notably, first-person singular pronouns when compared to healthy individuals. Deep learning paradigms are currently transforming what is possible in linguistic analysis. For instance, in sentiment analysis, machines are now capable of reaching near human recognition rates.
The aim of this study is to explore associations between language usage and mental state by using automated text analysis methodologies.  
Task In this thesis, the student(s) design and implement a robust method to classify and analyse transcripted audio recordings of a recently conducted depression study.
Utilises Word Embeddings, Attention Neural Networks , Recurrent Neural Networks.
Requirements Preliminary knowledge in Machine Learning and Natural Language Processing, Good programming skills (e.g. Python, C++).
Languages English or German.
Supervisor Lukas Stappen, M. Sc. (