Detecting depression from social networks using natural language processing

TAVCHIOSKI, ILIJA (author), Robnik Šikonja, Marko (mentor), Pollak, Senja (comentor)

The rapid technological advances in the past two decades have drastically affected our society's behavior, culture, and lifestyle. Social media, as one of the many products of this phenomenon, become an essential part of our lives as a tool for communication and expression. Social media become a popular choice for people to share information with the community, such as their thoughts and feelings on various matters in their lives. This especially can be observed in people with mental health issues, such as depression. They often prefer to express their strong feelings or ask for advice on social media. Social media posts also offer a possibility for automatically detecting signs of depression. In this thesis, we present the solution to this problem using natural language processing methods on two different data sets which are composed of posts from the social platforms Reddit and Twitter. We propose using large pre-trained models such as the BERT model and the use of transfer learning between the two data sets. We additionally improved the results by creating an ensemble of several combinations of transformer-based models pre-trained on different domains.

• Language: English
• Keywords: Natural Language Processing, Transformers, Depression Detection
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2022
• PID: 20.500.12556/RUL-140245
• COBISS.SI-ID: 123092995
• Publication date in RUL: 13.09.2022

TAVCHIOSKI, ILIJA, 2022, Detecting depression from social networks using natural language processing [online]. Bachelor’s thesis. [Accessed 11 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=140245

Secondary language

• Language: Slovenian
• Title: Detekcija depresije na družbenih omrežjih z metodami obdelave naravnega jezika

Abstract

Hiter tehnološki razvoj v zadnjih dveh desetletjih je močno vplival na vedenje, kulturo in življenjski slog družbe. Družbeni mediji, ki so eden izmed rezultatov tega fenomena, so postali ključen del naših življenj in pomembno orodje za komunikacijo in izražanje. Postali so priljubljen način, da ljudje s skupnostjo delijo informacije, na primer svoje misli in vprašanja o različnih zadevah svojega življenja. To lahko opazimo tudi pri osebah s težavami v duševnem zdravju, kot je depresija, ki družbene medije pogosto uporabljajo za izražanje svojih čustev ali tam prosijo za nasvete. Objave na družbenih omrežjih lahko služijo tudi za avtomatsko zaznavanje znakov depresije. V diplomskem delu predstavimo možne rešitve te naloge z uporabo metod za obdelavo naravnega jezika. Uporabimo dve različni zbirki podatkov, ki sta sestavljeni iz objav na družbenih platformah Reddit in Twitter. Predlagamo uporabo velikih prednaučenih jezikovnih modelov, kot je model BERT in uporabo učenja s prenosom znanja med zbirkami podatkov. Rezultate dodatno izboljšamo z uporabo ansamblov z različnimi kombinacijami modelov, ki temeljijo na transformerjih, prednaučenih na različnih domenah.

• Keywords: obdelava naravnega jezika, transformerji, zaznavanje depresije