Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
PLOUMIDI PANAGIOTA
Supervisors info:
Σταματόπουλος Παναγιώτης, Επίκουρος Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, ΕΚΠΑ
Original Title:
Suicide risk detection on social media using neural networks
Translated title:
Suicide risk detection on social media using neural networks
Summary:
According to World Health Organization records, approximately 280 million people suffer from depression and over 700.000 people die due to suicide while suicide is
the fourth leading cause of death among adolescents and young people. Whilst depression is a treatable condition, most people refuse to accept that they are affected and
therefore seek psychiatric help, due to social stigma associated with mental disorders,
especially in middle-income countries.
Social media platforms host people of all kinds of demographic groups and characteristics
and they thrive on young people. For most people social media set a safe space where
they can share thoughts and concerns, especially when they are covered by anonymity.
Platforms like Facebook and Instagram have created report options for such cases where
users can report a post that implies suicidal actions. It is of high importance that this procedure becomes automated, so that no users in need slip our attention and also depressive
tendencies can be predicted when it is still early. To resolve this problem, this thesis suggests a NN model that is trained on Reddit users’ posts and can reliably predict if a user
shows depressive or suicidal signs by examining his posts. The suggested NN is a hybrid
model that combines CNN and Bi-LSTM networks and also uses an attention mechanism
to optimise predictions.
Main subject category:
Technology - Computer science
Keywords:
artificial intelligence, neural networks, social media, emotion detection, nlp
