Exploration of Twitter Sentiments and Classification by using Deep CNN and Naive Bayes

Abstract

Sentiment evaluation of tweets help the enterprises to evaluate public emotion towards the activities or products associated with them. Most of the research targeted to obtain sentiment capabilities with the help of analyzing syntactic and lexical features which can be expressed through sentiment phrases, emoticons, exclamation marks etc. In the proposed paper we introduce a phrase embedding received by means of unsupervised learning(deep learning) on large twitter texts which uses contextual semantic relationships and co-occurrence statistical characteristics between words in tweets and also con- sider the emojis to categorise the emotions whether it is positive or negative by the use of Naive Bayes. In the preceding paper which used usnsupervised learning approach for classification, has an accuracy of 87% and supervised has an accuracy of 89%. According to our context, Naive Bayes has given an accuracy of 100% and CNN has given an accuracy of 100%. As compared to machine learning. It has a higher performance on the accuracy, precision and recall.