ANALISIS SENTIMEN ULASAN APLIKASI TIKTOK DI GOOGLE PLAY MENGGUNAKAN METODE SUPPORT VECTOR MACHINE (SVM) DAN ASOSIASI

Corona virus pandemic requires people to do activities from home so the number of internet usage in Indonesia has increased because information is carried out through social media. One of the popular social media in Indonesia is TikTok. However, the Tiktok’s popularity cannot be separated from the footsteps of TikTok in Indonesia which was blocked by government for committing many violations. Each application allows users to provide a review about the application. To find out the users TikTok’s sentiment, sentiment analysis was carried out to classify reviews into positive and negative sentiments. Classification is carried out using the Support Vector Machine (SVM) with kernel Radial Basis Function (RBF) method which is more effective classification algorithm and kernel function, seen from previous studies. The parameters used in the SVM gamma default 0.0004255 and the Cost (C) parameter experiment used is 0,01; 0,1; 1; 10; 100; 1000. The results can provide information that can be retrieved using the association method. The steps are scrapping data, data preprocessing, sentiment scoring, TF-IDF weighting, classifying using the SVM RBF kernel method and text association. Evaluation of the model using a confusion matrix with the value of accuracy and kappa. The greater the value of accuracy and kappa, the better the performance of the classification model. The review classification resulted in the best accuracy rate of 90.62% and the best kappa of 81.24% which means that it includes an almost perfect classification result. Based on the data association, positive reviews are given because users like and are comfortable with the current version of TikTok which contains funny videos on fyp. Meanwhile, negative reviews were given because the user failed to register and his account was blocked, so the user asked TikTok to continue to make improvements.

Multi-Dimensional Global Approximation Method Based Improved MARS

Global approximation for a complex “black-box” model (like a simulation model) with large domain or multi-dimensions can be applied in many fields such as parameter experiment, sensibility analysis, real-time simulation, and design/control optimization. For multi-dimensional global approximation, MARS (multi-variant adaptive regression splines) has unquestionable predominance over other common-used metamodel techniques. However, MARS has its own inevitable drawbacks which limit the range of its applications. This paper proposes a multi-dimensional global approximation method based improved MARS .Some tests and applications are given to prove the performance of the method.