Sentiment analysis of tweets by convolution neural network with L1 and L2 regularization

Abstract

Twitter data is one of the largest amounts of data where thousands of tweets are generated by the Twitter user. As this text is dynamic and huge so, we can consider it as a big data or a common example of Big data. The biggest challenge in the analysis of this big data is its improvement in the analysis. In this paper, there is an analysis by using semantic features like bigram, tri-gram and allow to learn by convolution neural network L1 and L2 regularization. Regularization is used to overcome the dropout and increase the training accuracy. In our experimental analysis, we demonstrated the effectiveness of a number of tweets in term of accuracy. In the result, we do not obtain any specific pattern but average improvement in the accuracy. For the analysis, we use 10 cross-validations and used to compare the outcome with max-entropy and SVM. Here we also analyze the effect of convolution layer on accuracy and time of execution.