Perbandingan Naïve Bayes dan Support Vector Machine untuk Klasifikasi Ulasan Pelanggan Indihome

Abstract

Abstrak IndiHome merupakan penyedia layanan internet dari PT. Telekomunikasi Indonesia, Tbk dengan jangkauan internet terluas di Indonesia. Kepuasan pelanggan menjadi salah satu hal yang harus diperhatikan dalam sebuah perusahaan, termasuk perusahaan IndiHome. Tingkat kepuasan layanan pelanggan IndiHome ini dapat dilihat dari ulasan pelanggan melalui media sosial Twitter. Penelitian ini membahas mengenai klasifikasi ulasan pelanggan IndiHome dengan mengaplikasikan tahapan penelitian CRISP-DM serta penerapan algoritme Naïve Bayes Classifier dan Support Vector Machine Kernel Linear. Data ulasan pelanggan diperoleh dari Twitter yang berjumlah 1000 tweet dengan menggunakan tools RapidMiner dan library R. Adapun tahapan preprocessing yang diterapkan yaitu cleansing, case folding, tokenizing, convert word, stopword, dan stemming. Hasil visualisasi data disajikan dalam bentuk word cloud yang dikategorikan berdasarkan opini positif dan negatif dari kata yang sering muncul. Hasil penelitian menunjukkan penerapan algoritme Support Vector Machine Kernel Linear lebih baik dibandingkan algoritme Naïve Bayes Classifier dengan nilai accuracy 82,11%, precision 76,44%, recall 88,01%, dan nilai AUC 0,909. Abstract IndiHome is an internet service provider from PT. Telekomunikasi Indonesia, Tbk with the widest internet coverage in Indonesia. Customer satisfaction is one of the things that must be considered in a company, including the IndiHome company. IndiHome's customer service satisfaction level can be seen from customer reviews via Twitter social media. This study discusses the classification of IndiHome customer reviews by applying the CRISP-DM research stages and the application of the Naïve Bayes Classifier algorithm and the Linear Support Vector Machine Kernel. Customer review data were obtained from Twitter, totaling 1000 tweets using the Rapid Miner and R library tools. The preprocessing stages applied were cleansing, case folding, tokenizing, word conversion, stopword, and stemming. The results of data visualization are presented in the form of a word cloud which is categorized based on positive and negative opinions of words that often appear. The results showed that the application of the Support Vector Machine Kernel Linear algorithm is better than the Naïve Bayes Classifier algorithm with an accuracy value of 82.11%, 76.44% precision, 88.01% recall, and an AUC value of 0.909.