ASPECT BASED SENTIMENT ANALYSIS IN TOKOPEDIA REVIEW USING LONG SHORT TERM MEMORY (LSTM)

M SATRIA YUDA UTAMA, Edi Winarko, Drs., M.Sc., Ph.D

2020 | Skripsi | S1 ILMU KOMPUTER

Nowadays, people are used to express their experience of buying things. People express their experience in the review section in marketplace where they bought their things. The large number of review in review section may helps prospective buyer to choose which seller they will choose to buy. However, the prospective buyer may confused the relevant information, since there are a large number of reviews. In this research, the author propose aspect based sentiment analysis that automatically give what aspect that people tell in a review sentence followed with the sentiment of aspect in there. In this research, the author explore the usage of Sequence-to-Sequence (Long Short Term Memory) LSTM to detect the aspect in review sentence, and explore the usage of LSTM to classify the sentiment of the sentence review with each aspect. The proposed model of aspect detection give accuracy of all model is 74.7%, recall value is 78.3%, precision value is 80.5% and the F1-Score 79.4%. and for the aspect sentiment classification model give 91.07%.

Nowadays, people are used to express their experience of buying things. People express their experience in the review section in marketplace where they bought their things. The large number of review in review section may helps prospective buyer to choose which seller they will choose to buy. However, the prospective buyer may confused the relevant information, since there are a large number of reviews. In this research, the author propose aspect based sentiment analysis that automatically give what aspect that people tell in a review sentence followed with the sentiment of aspect in there. In this research, the author explore the usage of Sequence-to-Sequence (Long Short Term Memory) LSTM to detect the aspect in review sentence, and explore the usage of LSTM to classify the sentiment of the sentence review with each aspect. The proposed model of aspect detection give accuracy of all model is 74.7%, recall value is 78.3%, precision value is 80.5% and the F1-Score 79.4%. and for the aspect sentiment classification model give 91.07%.

Kata Kunci : Aspect Based Sentiment Analysis, Sequence-to-Sequence, LSTM, Aspect Detection, Sentiment Analysis