ANALISIS KLASIFIKASI TOPIK MENGGUNAKAN METODE NAÏVE BAYES CLASSIFIER, NAÏVE BAYES MULTINOMIAL CLASSIFIER, DAN MAXIMUM ENTROPY PADA ARTIKEL BERITA; (TOPICS CLASSIFICATION ANALYSIS USING NAÏVE BAYES CLASSIFIER, MULTINOMIAL NAÏVE BAYES CLASSIFIER, AND MAXIMUM ENTROPY FOR NEWS ARTICLES)

MASITHOH, NURUL, Herni Utami

2016 | Skripsi | FMIPA

The rapid growth of website based communication nowdays gives easier access to obtain more information. The demand for better accuary and faster information is increasing now due to the modern lifestyle. Most of the data are provided in textual verse aren’t presented in any structured patterns. Statistics is playing a prominent role in this part, especially in completing textual data. Text mining is an analitical system that used in textual data. This method which are often applied in analytical process are text classification and topic classification. Based on result of topic classification, analyst can decide the topic type. The topic itself is concluded based on the document characteristic. This paper will discuss about Naïve Bayes Classifier, Naïve Bayes Multinomial Classifier, dan Maximum Entropy. Those methode will be applied to define the topic class based on multiclasses: ball topic class, health, economy, and tavel. Naïve Bayes Classifier and Naïve Byaes Multinomial Classifier are classifying methods that use Bayesian rules. The rules are including prior probability and conditional probability based on the appearance of words, for Naïve Bayes Classifier and the frequency of words appearance for Naïve Bayes Multinomial Classifier in each document training. While Maximum Entropy uses approximate information which is included in the document. This information is optimized so that the classification becomes more accurate. That value will be applied to decide the topic class, by looking at the maximum a posterior in each class. Classified data would be counted in order to find the accuracy level. Based on the comparision of approximate values from those methods, the leveling of accuracy. The highest level is Maximum Entropy (99,31%), then Naïve Bayes Classifier (98,82%), and the lowest level is Naïve Bayes Multinomial Classifier (97,39%). This research was applied on 1440 news articles. Each articles underwent 87 trials. Not only comparing those three methods, but also analyzing the spect of token data usage, number of data usage, and the similary number of data on value accuary. The conclusion is those three aspects don’t have influence the data accuary in each method.

Kata Kunci : Topic classification, text mining, Bayesian rules, Naïve Bayes Classifier, Naïve Bayes Multinomial Classifier, Maximum Entropy.