Sentiment Analysis of Bengali Reviews for Data and Knowledge Engineering: A Bengali Language Processing Approach

Abstract

Opinion mining is very much attractive field inmachind learning system as it is very much needed for natural language processing. The opinion mining of Bengali written English word has been done successfully using four different classifiers—support vector machine, naive Bayes, logistic regression and random forest. For the work data set was extracted from local online shops using pursehub. The work was done with vital steps—data preparation, classifying reviews according to sentiment score and evaluate the system in all steps. The F1 score was obtained 85.25%, 88.12%, 88.12%, 82.43% for naïve Bayes, logistic regression, SVM, random forest, respectively. The accuracy score was obtained 85.31%, 88.05%, 88.11%, 81.82% for naïve Bayes, logistic regression, SVM, random forest, respectively. The precision score was obtained 85.56%, 88.54%, 87.59%, 79.14% for naïve Bayes, logistic regression, SVM, random forest, respectively. The recall score was obtained 84.95%, 88.72%, 88.80%, 85.30% for naïve Bayes, logistic regression, SVM, random forest, respectively.