ASPECT BASED SENTIMENT ANALYSIS USING CONVOLUTION NEURAL NETWORK AND BIDIRECTIONAL LONG SHORT-TERM MEMORY

• Main Author: CAHYADI (NIM : 13514035), ALSON
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25409
• Institution: Institut Teknologi Bandung

<p align="justify"> Sentiment analysis is able to replace survey as a means for executives to make long, medium, and short term decisions. Prior research of aspect-based sentiment analysis (ASBA) Indonesian restaurant review has been enhanced by emphasizing the use of diverse features. This final project adapts the best research method of SemEval 2016 task 5 for English to improve the performance of aspect-based sentiment analysis for Indonesian restaurant reviews. In SemEval 2016 task 5, there are three slots to do for ASBA namely aspect category (slot 1), expression target target (slot 2), and sentence polarity (slot 3). This final project produces three learning systems to handle the new slot SemEval 2016 task 5. The model used for slot 1 is a binary classifier that each consists of artificial neural networks with a probability feature of Neural Convolutional Networks. For slot 2 models are made sequential by using Conditional Random Field (CRF) with probability feature of Long-Term Memory Long Term (BLSTM), POS tag, word, and cluster id. Slot 3 is equipped with a CNN model. Test and train data used in this final project is test data used in ASBA research to see Indonesian restaurant with food, service, price, and place category. The test data consists of 383 Sentences of train data comprising 992 sentence reviews from the TripAdvisor site. F1-measure of 1, 2, 3, and universal slot slots are 0.8704, 0.7873, 0.8992, and 0.7637. <p align="justify">