Tweet Sentiment Classification Using an Ensemble of Unsupervised Dictionary Based Classifier and Deep Convolutional Neural Network

Abstract

Social media has given a very good ground for internet users for sharing their opinion and reviews about various subjects. Large amount of data is generated by various social networking sites. Analysis of this data can be very useful in decision support. Twitter is a very good platform for sentiment analysis.  Opinion mining or sentiment analysis is used to predict the opinion of people on different issues which tell how people feel about that issue. Vast range of algorithms is used for opinion mining. In this paper Deep Convolution Neural Network with dictionary classifier is used for Twitter sentiment classification.  Sentiment score extracted from dictionary classifier is added in word of embeddings. The purpose is more accurate prediction of opinion.  Keras and Tenserflow are used for the implementation of model on manually labeled Real time tweets.  The results show that CNN model performed best as compared to other machine learning classifiers. Also CNN with sentiment score further improved the performance and the model classified the tweets with an accuracy of   0.9325.