A Comparative Study of a New Customized Bert for Sentiment Analysis

AbstractThis paper presents a new customized Bert method based sentiment analysis classification. The blank space removal method, stop word removal, and stemming methods were used in the data pre-processing step. Then tokens are extracted in tokenization step. Standard lion algorithm is used to optimize the weight and bias of the proposed Bert algorithm. Many comparisons were done in order to analyze the performance of the new proposed Bert model. Results obtained showed that the proposed model outperforms other classical models in term of accuracy and performance other metrics.KeywordsBertTwitter sentiment analysisDeep learningLion algorithmOptimization