Effectiveness of the News Text Classification Test Using the Naïve Bayes’ Classification Text Mining Method

Abstract

This study used a quantitative approach created by the author to describe the accuracy of the classification machine learning for news text. In this report we compare the results of the accuracy values obtained using the Naïve Bayes method with other methods to see the effectiveness of the method used. The resulting accuracy value has not reached its maximum and thus it could still be restructured and re-evaluated into a better model. In this case the writer tried to increase the precision value produced in order to make this machine able to predict news that contains sarcasm through modification of the set threshold value. After the threshold value was changed to 0.3, the accuracy value decreased to 61% but the precision value increased to 77% and the error value in the prediction of false positive headlines also increased significantly and only produced 89 errors in predicting sarcasm. The ROC Curve test showed that this machine learning model could still be improved by trying other text preprocessing methods such as the bigrams, tidytext, lemmatization methods, so that the machine will become smarter at predicting the resulting vectors and increase the value of precision and accuracy obtained.