Sentiment Analysis Performance Value Optimization Using Hyperparamater Tunning With Grid Search On Shopee App Reviews

Abstract

The rapid development of technology today has provided convenience for us in today's civilization. One of these developments is the invention of the internet due to high internet penetration and rapid growth in mobile usage, online shopping has increased tremendously. This online shopping is now often referred to as e-commerce. E-commerce is one of the trade models that has been widened under the effect of extensive use of technology. Specifically, e-commerce refers to the usage of the Internet or other networks. Shopee is one of the popular marketplaces in Indonesia that has the highest number of visitors of 129 million per month and can be downloaded on the Google Play Store. Play Store itself has several features such as Reviews that can allow users to give opinions. All complaints and opinions from shopee users can be channeled into this feature. With this a research aims to optimize the performance value of sentiment analysis with the Term Frequency-Inverse Document Frequency (TF-IDF) method and Hyperparameter Tuning with Gridsearch for the Shopee application on the Google Play Store. Based on research the reviews resulting in 3000 data where 2015 user data is positive and  985 data is negative. Testing data was split by a ratio of 90:10 for 300 data test in each classification model to find the accuracy score. With hyperparameter tuning using gridsearch we can see the result of each accuracy score of KNN, DCT, RF, and LR is increasing from 0.73 to 0.77, 0.823 to 0.826, 0.856 to 0.87, and 0.856 to 0.866. This indicated that among the machine learning model that had been tuning using gridsearch, KNN is the one that highly increased.