Computational Statistical Analysis of the Language of Religious Discussions on Internet Forums

Analiza religijskih pitanja je uvijek vrlo kontroverzna. Religija kod svakog čovjeka predstavlja nešto osobno, a objektivno sagledati religiju je vrlo zahtjevna stvar. Cilj rada je bio analizirati tekstove religijskih rasprava uz pomoć statističke analize. Skup podataka korišten u radu smo preuzeli s društvene mreže Reddit. Model smo izgradili koristeći frekvencije riječi i značajke LIWC. Isprobali smo dva modela strojnog učenja, logističku regresiju i stroj potpornih vektora za predviđanje religije. Ostvarili smo rezultate koji su zadovoljavajući te s točnošću od otprilike 65% predviđaju religiju korisnika na temelju njegovih ili njezinih komentara.Analysis of religion questions is always very controversial. Religion is something personal for every person, and to objectively consider religion is a very demanding thing. The aim of the thesis was to analyze the texts of religious discussions through statistical data analysis. The data set used in this work was downloaded from the Reddit social network. We built the model using word frequencies and LIWC features. We tested two machine learning models, logistic regression and the support vectors machine for predicting religion. We have achieved satisfactory results and with an accuracy of approximately 65% predict the user's religion based on his or her comments