COMPARISON OF LSTM VARIANTS AS A SENTIMENT PREDICTION OF MARVEL CINEMATIC UNIVERSE’S MOVIE REVIEW

FAKHRI, MUHAMMAD AMMAR (2020) COMPARISON OF LSTM VARIANTS AS A SENTIMENT PREDICTION OF MARVEL CINEMATIC UNIVERSE’S MOVIE REVIEW. S1 thesis, Universitas Mercu Buana Jakarta.

Text (JURNAL MAHASISWA) Yudisium M. Ammar Fakhri (41516010027).pdf Restricted to Registered users only Download (4MB)

Abstract

Sentiment Analysis is a field that studies people’s opinions expressed in a text format. Long Short-Term Memory (LSTM) is often used to solve various cases in the Natural Language field because of their decent performance and that they can overcome the problems that occurred in Recurrent Neural Network. This study will be focused on a comparison between 4 variants of LSTM (Basic LSTM, Bidirectional LSTM, Deep LSTM, and Deep Bidirectional LSTM) to predict the sentiment from reviews of each movie review in the 3rd phase of Marvel Cinematic Universe. We find out that Bidirectional LSTM can get the best validation accuracy with a maximum of 84.73% and an average of 84.36%, while Deep Bidirectional LSTM has the smallest epochs to be trained in an average of 90.4 epochs Key words: Sentiment analysis, movie review, RNN, LSTM Analisis Sentimen adalah bidang yang mempelajari opini masyarakat yang dituangkan ke dalam bentuk teks. Long Short-Term Memory (LSTM) sering digunakan untuk menyelesaikan berbagai masalah pada bidang Natural Language karena dapat memberikan performa yang baik dan mampu mengatasi masalah yang ada pada RNN. Penelitian kali ini akan berfokus pada perbandingan empat varian dari LSTM, yaitu Basic LSTM, Bidirectional LSTM, Deep LSTM, dan Deep Bidirectional LSTM untuk memprediksi sentimen dari ulasan film-film yang ada pada fase ke-3 Marvel Cinematic Universe. Bidirectional LSTM dapat memberikan performa akurasi terbaik dengan maksimal akurasi sebesar 84,36%, sedangkan Deep Bidirectional LSTM dapat melakukan training lebih cepat dibanding algoritma lain dengan rata-rata epoch sebesar 90,4. Kata kunci: Analisis sentimen, ulasan film, RNN, LSTM

Actions (login required)

View Item