Comparative Evaluation of Sentiment Analysis Methods

Abstract

Sentiment analysis is an on-going field of research within the realm of Natural Language Processing, in which we wish to accurately assess the sentiment of an author on a given topic. Within this thesis project I construct a rule-based system for sentiment analysis specific to the domain of hard rock and heavy metal music album reviews. I then compare it to the performance of other approaches to the task, such as the use of a neural network, and analyse the strengths and weaknesses of these differing approaches. Ultimately the neural network, with sufficient training, produces the best results for this task, and I go on to outline possible improvements that could be made to the rule-based system in further efforts to maximise its potential.