Modeli strojnog učenja za otkrivanje psihičkih bolesti i poremećaja autora teksta

master's thesis

• Title (english): Machine Learning Models for Detecting Psychological Illnesses and Disorders of Text Authors
• Author: Viktor Golem
• Mentor: Jan Šnajder (mentor)
• Committee member: Jan Šnajder (predsjednik povjerenstva)
• Committee member: Bojana Dalbelo-Bašić (član povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-08, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Pravodobno prepoznavanje bolesti gotovo je jednako važno kao i povećanje učinkovitosti njihovog liječenja. Rano se prepoznavanje uglavnom svede na odlazak kod liječnika potaknut pojavom prvih simptoma. Kod psihičkih bolesti i poremećaja situacija je značajno drukčija - ne samo zato što osim promjena u ponašanju pojedinca drugih simptoma gotovo da i nema, već zato što postoje istraživanja koja upućuju na to da se u govoru i pismu osoba koje pate od psihičkih poremećaja i bolesti mogu uočiti određeni uzorci i pravilnosti. Primjena računala ili, preciznije, strojnog učenja u prepoznavanju psihičkih problema autora teksta mogla bi biti iznimno korisna. Tada je očito nužno prikupiti dovoljnu količinu podataka, što je najjednostavnije postići koristeći brojne objave korisnika društvenih mreža. Autori skupa podataka SHMD napravili su upravo to - prikupili su objave korisnika \textit{Reddita} i označili ih poremećajima, odnosno bolestima, od kojih ti isti korisnici pate (ili su, u slučaju da su zdravi, označeni kao dio kontrolne skupine). Koristeći navedeni skup podataka kroz ovaj je rad istražena primjenjivost različitih modela temeljenih na LSTM-u treniranih na različite načine. Prvi način se pokazao užasno sporim i nedovoljno kvalitetnim, zbog čega nije bilo dovoljno vremena za detaljnije proučavanje drugih pristupa koji su davali obećavajuće rezultate.

Abstract (english)

Being able to detect diseases early enough to facilitate their treatment is a task almost as important as increasing the efficiency of the treatment itself. But early detection mostly relies on appearance of symptoms that might prompt the person to talk to a doctor. When it comes to psychological disorders and illnesses, things are different - not only because they do not exhibit symptoms other than changes in behaviour, but because research has shown that certain patterns in speech and text produced by people suffering from certain disorders can be captured. Using computers, more specifically, machine learning to detect mental health issues of text authors might be quite beneficial. To do that, enough data has to be collected, so it is only natural to resort to social networks where users choose to share different content on daily basis. The authors of SHMD dataset aimed to do exactly that - gather user posts from Reddit and label them with the disorder the user suffers from (or assign the user the control group label if healthy). In this work using that dataset several LSTM-based models in different setups were trained, the first being rather time consuming without achieving satisfactory results, leaving less time to explore the more promising approach in the second setup.

• Keywords: strojno učenje; obrada prirodnog jezika; psihički poremećaji i psihičke bolesti; LSTM; multitask learning; word embeddings
• Keywords (english): machine learning; NLP; mental disorders and illnesses; LSTM; multitask learning; word embeddings
• Language: english
• URN:NBN: urn:nbn:hr:168:023166
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 18:09:42