Abstract:
The proliferation of social media enables the public to express their views and
perceptions readily online. Twitter is one such platform that helps in obtaining a
huge amount of textual data and performing useful analysis. Sentiment
Classification is one such analysis undertaken to gain insights into public opinion
on a certain topic. Although this has been prevalently done using many
approaches, the limitations still exist in non-English languages. This study aims
to compare the use of the lexical-based approach and machine learning-based
approach for classifying the Tamil tweets based on their sentiment. Twitter API
was used to perform twitter scraping that resulted in 45852 tweets in total. 300
random tweets were then classified to their respective sentiments by subject
experts in the field, this annotated data was used as ground truth and 06
underlying studies were performed on the processed and cleaned data. Four
machine learning algorithms (Support Vector Machine, eXtreme Gradient
Boosting, Random Forest, and Gaussian Naïve Bayes) and two lexical-based
analyzers (VADER and TextBlob) were used for this comparative analysis. The
results suggested that the machine learning algorithms performed extremely well
where the Support Vector Machine secured the best performance score of all. This
study serves as empirical evidence for those interested in performing sentiment
analysis on Tamil language tweets.
