Optimalisasi Model Klasifikasi Sentimen Netizen Terhadap Merek Tas Luar Negeri

Abstract

Abstract
 Research on text mining has grown more than ever in various sectors. Public figures have also grown in interest towards the field and have the tendency to get to know more about consumers’ perceptions toward relevant goods and the reputation of an individual in social media. Sentiment analysis is a state-of-the-art technique that can be utilized to evaluate such trends or general views, for instance the reputation of a fashion brand. The dataset is built upon the crawled tweets that are relevant with the required topics which have the purpose to analyze the preferred fashion brand of the public. This study shows that the public leads to a positive notion toward foreign bag brands. The algorithms that are being compared includes Logistic Regression, Multinomial Naïve Bayes, Decision Tree, K-Nearest Neighbors, Random Forest, and Support Vector Machine. Support Vector Machine provides the best model which reaches 69% in accuracy. The Synthetic Minority Oversampling Technique (SMOTE) was also conducted to improve the model. Result shows that the Support Vector Machine model has successfully increased its accuracy by 13%, reaching an accuracy of 82%.
 Keywords: Sentiment Analysis, Brand, Machine Learning, Classification, SMOTE
 Abstrak
 Penelitian mengenai text mining telah mengalami peningkatan dibanding sebelumnya di dalam berbagai sektor. Figur publik juga semakin tertarik terhadap bidang tersebut dan memiliki kecenderungan untuk mengetahui lebih banyak mengenai persepsi konsumen terhadap suatu barang dan mengenai reputasi seseorang di media sosial. Sentimen analisis merupakan sebuah teknik state-of-the-art yang dapat digunakan untuk mengevaluasi suatu tren atau pandangan umum mengenai suatu hal, misalnya reputasi sebuah merek fashion. Sumber himpunan data yang digunakan pada penelitian ini dibuat berdasarkan crawling tweet yang relevan dengan topik yang dibutuhkan, yang bertujuan untuk menganalisis merek fashion yang disukai oleh masyarakat. Penelitian ini menunjukkan bahwa persepsi masyarakat mengarah pada persepsi positif terhadap merek tas luar negeri. Pada penelitian ini, beberapa algoritma digunakan sebagai perbandingan, antara lain Logistic Regression, Multinomial Naïve Bayes, Decision Tree, K-Nearest Neighbors, Random Forest, dan Support Vector Machine. Hasil pengujian model menunjukkan algoritma Support Vector Machine memiliki performa terbaik dengan accuracy sebesar 69%. Kemudian digunakan teknik Synthetic Minority Oversampling Technique (SMOTE) untuk meningkatkan performa dari model. Hasil menunjukkan bahwa model algoritma Support Vector Machine telah berhasil ditingkatkan dengan accuracy sebesar 13%, mencapai accuracy sebesar 82%.
 Kata kunci: Sentimen Analisis, Merek, Pembelajaran Mesin, Klasifikasi, SMOTE

Analysis of Cyberbullying on Social Media Using A Comparison of Naïve Bayes, Random Forest, and SVM Algorithms

“…Next, the tokenization and normalization process is carried out, which aims to separate each word that forms a sentence into token pieces. [18] and normalize each word, including abbreviations or typos, so that they will be corrected to be more structured and can be further processed based on the prepared dictionary.…”