A Comparative Study of Naive Bayes, Vader, and TextBlob Methods in Sentiment Analysis of ShopeeFood on Twitter

Twitter API tweets were utilized to analyze sentiments surrounding ShopeeFood using its algorithmic attachment. A contended sample of 2,500 tweets was gathered for Shapshot-1 in sample focus and was later cleaned and translated into English. The methods employed for the analysis include TextBlob, VADER, and Naïve Bayes classifiers. The analysis reconsolidated, yet again, that tweets, which, by and large, had neutral sentiments attached to them, as confirmed by Naïve Bayes out of 83 per cent accuracy attained. VADER's classification resulted in 85.08% of tweets being categorized as neutral, positive 9.4%, and negative 5.52%. All three constructs captured presented similar results, but the Naive Bayes model proved to be more favourable in terms of sentiment classification; despite such successes with VADER and TextBlob, feature selection and the changes from the translation left them a flaw within the analysis. These problems highlight the challenges posed by social media data, which is rife with casual language, slang, and emoticons. To overcome these challenges, future work should focus on employing neural network techniques that would bolster performance for sentiment classification on large corpora. Practices such as the collection of social media opinion sentiment within the pre-processing stages need more focus. More sophisticated models and advanced pre-processing methods can yield more fine-grained sentiment and opinion expressions on Twitter.