Sentiment analysis of potentially discriminatory messages in media commentary

Abstract

The main aim of this dissertation focuses on sentiment analysis. It aims at training different classifiers to label data as positive or negative. This data is obtained from a corpus of pre-labelled data of local online news comments called ‘C.O.N.T.A.C.T.’ which is a corpus of comments made as a reaction to Maltese news portal stories pertaining to migration and LGBTQI issues. These issues bring about various types of reactions which can either be positive or else negative. Four experiments were administered based on classifiers which used different approaches. The four experiments were devised using a machine-learning based approach, for experiments 1, 2 and 4, and a dictionary-based approach of sentiment analysis using Python code for experiment 3. A comparison exercise was made between the pre-labelled data obtained from ‘C.O.N.T.A.C.T.’ and the results obtained after administering the experiments. Another comparison exercise was made between the data results of each experiment so as to deduct which approach had the highest accuracy rate.