Twitter Sentiment Analysis Based Classification Model incorporating Multinominal Naive Bayes Classifiers

The study is a small-scale effort to categorise positive and negative attitudes from Twitter comments. Social media makes it incredibly simple to disseminate hate speech, racist statements, and comments these days. Various social media sites have begun utilising Natural Language Processing (NLP) to identify and flag remarks that are associated with bigotry and hatred. This project uses a collection of actual tweets from Twitter to perform a sentiment analysis on the platform. There are 200000 observations in the dataset. Stopwords, punctuation, and Twitter user handles are all present in the tweets itself. After pre-processing the data to exclude these undesired features, the tweets are count vectorised and used to train a Naïve Bayes Classifier (NBC), whose performance is assessed using the F1 score, confusion matrix, recall score, along with precision score. The existing method of sentiment analysis includes assumption that the feature in the text are all independent, but in reality it is not so which leads to incorrect sentiment prediction. Along with this it also results Naïve Bayes classifiers failing to detect sarcasm the tweets being slang, Naïve Bayes classifiers fails to classifies abstract complete information from them, and it also fails to adapt well to evolving trends in language. To overcome these limitations a hybrid Naïve Bayes classifiers in the studies and a lot of previous problems have been eliminated. Prior they got 57% accuracy level in results and This paper shows 87% accuracy level for tweets. The assessment conclude with insights into future research directions, emphasizing the need for improving accuracy level.