THESIS
2017
1 volume (various pagings) : illustrations (some color) ; 30 cm
Abstract
This thesis lies at the crossroads of affective computing, psychology, and entrepreneurship. The
motivation for this thesis was personal as well as societal. Mental health issues are increasingly
common in the developed and developing world alike. However, access to care is often too
expensive or stigmatised. Computer platforms show great potential in alleviating this problem.
This thesis presents such a platform and its modules, as well as research that can help improve
virtual agents' performance.
This work describes Automatic Personality Prediction from audio, text, and video, using
end-to-end deep learning architectures such as Convolutional Neural Networks. We are
interested in combining these three modalities to improve performance and in exploring
how relevant each modal...[
Read more ]
This thesis lies at the crossroads of affective computing, psychology, and entrepreneurship. The
motivation for this thesis was personal as well as societal. Mental health issues are increasingly
common in the developed and developing world alike. However, access to care is often too
expensive or stigmatised. Computer platforms show great potential in alleviating this problem.
This thesis presents such a platform and its modules, as well as research that can help improve
virtual agents' performance.
This work describes Automatic Personality Prediction from audio, text, and video, using
end-to-end deep learning architectures such as Convolutional Neural Networks. We are
interested in combining these three modalities to improve performance and in exploring
how relevant each modality is. Additionally, for the case of audio analysis, we look at
multilingual approaches to personality prediction. We find that training on combined datasets
in different languages can improve the performance for each of these languages. We also
show that languages share common low-level features, and that mid-level features tend to be
language specific. The research is a significant contribution to the field of Automatic Personality
Prediction for the completeness of the models, insights into multilingual personality prediction,
and interpretability of trained models.
Furthermore, this thesis describes Project Nora, a hypothetical company which is building a
self-help platform called 'Nora the Virtual Therapist', with an interface in the form of a virtual
agent. Users can get initial screenings for mental health, guided deep breathing sessions, or
track their mood over time. All research efforts described in this thesis have been included in
the system. Nora is aware of user stress and personality, and adapts to it for a personalised
experience. Additionally, this thesis contains a business plan to explore the commercial viability
of Project Nora. We discuss market trends and the competitive environment, as well as a business model that makes sense for a project such as Nora.
Post a Comment