Abstract:
Sentiment Analysis is the study of classifying a given text based on its sentiment
(positive/ negative polarity) of the expression. Sentiment analysis is being widely used
to analyse the public opinion towards a given entity. Today in Web 2.0, social media
is a popular platform to express one’s opinions and beliefs. Therefore, researchers are
keen on investigating how social media sentiment analysis can be improved to benefit
interested entities. Most of the sentiment analysis research has been conducted on
identifying the polarity (i.e.: positive, negative or neutral) and emotions (i.e.:
happiness, sadness, disgust, anger, fear and surprise).
Comparatively, less focus has been given to study how expressions can be classified
based on psychological aspects of attitude. The objective of the proposed research is
to move beyond the mere polarity, and to investigate whether we can get an in-depth
understanding of the expressed attitude. For this, we have used the ABC (Affective,
Behavioural and Cognitive) model of attitude introduced in consumer psychology.
In this research a new dataset was compiled by extracting Tweets on a specific topic
and manually annotating them based on the attitude by domain experts. This research
discusses how existing tools and technologies of Sentiment Analysis can be applied
for this problem domain. Various preprocessing and feature extraction techniques were
evaluated against a set of machine learning algorithms including Ensemble and Deep
Learning models. Additionally, this research aims to contribute to reduce the gap
between machine learning and consumer psychology and thereby proving the
possibility of applying machine learning across different domains.
