The Comparison RNN and Maximum Entropy on Aspect-Based Sentiment Analysis of Gojek Application

Nowadays, mobile applications can help a person to carry out daily activities. The use of mobile applications is also increasingly in demand by the public. One of the most popular online transportation applications in Indonesia is Gojek, with the top level of the most downloads in Indonesia. However, Gojek also experienced a significant decline from the previous download results. This is used as sentiment analysis by the author to find out how users rated Gojek application reviews from various points of view. This research compares two methods, namely Maximum Entropy and Recurrent Neural Network (RNN) using Chi-Square as feature selection and TF-IDF as feature extraction for each aspect of Availability, System, Comfort, and Transaction. As for the results of user analysis of four aspects with positive and negative sentiment, it is carried out with a 70:30 comparison ratio because it gets a better accuracy result value. The results show that the RNN method gets a better accuracy value than the Maximum Entropy method, with an accuracy value in the accessibility aspect of 90%, system aspect of 89%, comfort aspect of 80%, and comfort aspect of 80%.