Abstract:
Contemplating the fact that business enterprises and individuals, globally, benefit from
social media to gain profits, this dissertation discusses the importance and impact of
analyzing mechanisms of consumer responses on social media to maximize the gained
profit and take befitting future decisions while reducing the risk of cyber-crimes and
terrorism. The main objective is applying sentiment analysis techniques on the collected
responses to analyze them in order to identify the emotions behind. This study focuses
on finding out the best model for emotion categorization of the collected responses
using the sentiment analysis techniques.
The collected responses are preprocessed and undergo several techniques to achieve
data normalization which enable fast and effortless querying. Cleansed data is analyzed
in order to place the comments under six main emotion categories related to sentiment
analysis. This analysis mainly consists of feature extraction and training the models.
The two built-in methods available in scikit learn toolkit, namely, TfidfVectorizer() and
CountVectorizer() are used for the feature extraction using bi-gram features and those
extracted features are used to train the selected four (04) classification models in order
to find out the best performing model with the two (02) feature types extracted.
Subsequently, this study will investigate and compare different features for the different
classifiers when categorizing emotions on social media. Then record the accuracy of
each model with each feature type and select the best performing model with the
accuracy of 80% with a 0.1909 train-test accuracy difference
