ELMAN RECURRENT NEURAL NETWORK FOR ASPECT BASED SENTIMENT ANALYSIS

Author

Nelly Indriani Widiastuti, M.T.

Abstrak

Aspect based sentiment analysis (ABSA) is one of the domains of opinion mining cases which aims to detect the polarity of written text based on certain aspects. The purpose of this study was to determine the accuracy value of Elman RNN in the case of ABSA. The method used is divided into two main processes, namely pre-process and sentiment detection. Before conducting the training, the input data in the form of restaurant reviews in Indonesian went through the preprocessing process. In the data review, case folding, filtering, word normalization, tokenization, stop word removal, the addition of token aspects (price, taste, atmosphere) was carried out, building a word dictionary, and forming one-hot encoding. In the polarity detection process, training and testing use the ERNN algorithm. The data used were 1584 sentences of Indonesian restaurant reviews and were tested on 422 data. Based on the test results, Elman RNN got the best accuracy of 81.22% and an f1 score of 82.78%. For the social analytic system developer, these results show evidence that the ERNN is promising to be used in detecting the polarity of a restaurant review.