Emotion Detection of Twitter Post using Multinomial Naive Bayes

In research fields, emotion analysis and opinion mining using data from different platforms are up burning field. In this paper, we tried to represent sentiment of twitter data on core text. But tweets can only be in 140 characters, with lots of noise. Tweets contain few words which is in short forms, ambiguous and noisy, so it is hard to figure out the user's sentiments. So, it becomes very difficult to have the right opinion with these noise and short forms of words. The main job is to preprocess the data and then extract the features from there. But preprocessing demands, different theories, methods, steps which always varies. Our goal is to improve the outcomes using Naive Bayes classifier and an almost a good trained data set. Finally, we have our average accuracy for happy class 60%, surprise class 61%, relief class it is 71% and worry class has the highest 81%, by using unigram model for preprocessing. On the other hand, using unigram with POS tag model we have average accuracy of 63% same for happy and surprise class, 72% for relief and 83% for worry class.