ANALISIS SENTIMEN TERHADAP ULASAN NOVEL REMAJA DENGAN MENGGUNAKAN NEURAL NETWORK

• Main Author: Syifa Widodo, Azzahra
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/65247
• Institution: Institut Teknologi Bandung

Sentiment analysis is a field of data science that aims to identify and extract opinions and attitudes that are expressed in written text. There are three levels of sentiment analysis: document level, sentence level, and aspect level. Document-level sentiment analysis is the most widely studied level. Sentiment analysis at this level has mainly been carried out using machine learning methods, such as the Naive Bayes classifier and support vector machines. However, it is becoming increasingly popular nowadays to utilise neural networks for sentiment analysis because these methods are able to extract patterns and dependencies between words in a piece of text. This Final Project aims to build several neural network models using Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) for sentiment analysis of young-adult novel reviews. We build five models, namely the CNN, LSTM, GRU, CNN-LSTM, and CNN-GRU models. We train and evaluate each of these models to predict the sentiment of a review text by varying the lengths of the text, namely at length 100 words, 125 words, and 150 words. The GRU model produces the best result at length 100 words with 86.75% accuracy, however the CNN-GRU model produces the best results at both length 125 and 150 words with 85.32% and 85.36% accuracy respectively.