ONLINE NEWS CLASSIFICATION USING NAÏVE BAYES CLASSIFIER WITH MUTUAL INFORMATION FOR FEATURE SELECTION

Abstract

The number of online news documents can reach billion documents. Therefore, the grouping of news documents required to facilitate a editorial staff to input and categorize news by its categories.
This paper aim to classify online news using Naive Bayes Classifier with Mutual Information for feature selection that aims to determine the accuracy from combination of this methods in the classification of online news documents, so grouping of online news documents can be classified automatically and achieve more accurate for classification model. Data is divided into training and testing data. Data in August, September and October 2016 was used for training data. For testing data, 65 documents was used that located in November. The best results of this methods are 80% of accuracy, 94.28% of precision, 79.68% of recall and 85.08% of f-measure for Multivariate Bernoulli without feature selection. Then the best results of classification model using Mutual Information for feature selection achieved in Multivariate Bernoulli model with 70% of accuracy, 89.11% of precision, 69.76% of recall and 78.04% of f-measure with the word’s efficiency rate until 52% than before using feature selection. In other hand, the results of Multinomial Naïve Bayes without feature selection are 41.67% of accuracy, 75.68% of precision, 41.90% of recall and 48.13% of f-measure, for the results of Multinomial Naïve Bayes model using feature selection are 10% of accuracy, 33.33% of precision, 9.40% of recall and 14.35% of f-measure.