HOLDER AND TARGET IDENTIFICATION ON OPINION TEXT USING DEEP NEURAL NETWORKS

• Main Author: Mirza Maulana Ikhsan, Moh.
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/68615
• Institution: Institut Teknologi Bandung

The development of social media has made everyone to be able to express their opinion on the internet. Therefore, various techniques have been developed to extract any information contained in opinion texts. Opinion Role Labeling (ORL) aims to identify the opinion holder and opinion target within opinion text. This final project aims to build a deep learning model to identify opinion holders and opinion targets in an opinion text. This final project uses the MPQA 2.0 (Multi Purpose Question Answering) corpus which consists of a various news in English language. The final project focuses on experimenting with various deep learning architectures in the research baseline model by Quan, et al., (2019) and knowing the performance of each model produced using a 10-fold cross validation scheme. After that the performance results of each model are then compared to find out which model has the best performance. Based on experiments, the results of using Convolutional Neural Network (CNN) architecture for character level feature extraction can increase the performance of the BERT-BiLSTM CRF baseline model by 3%. In addition the use of opinion expression feature in the model can significantly increase the performance of the baseline model by 20%. Therefore, the BERT-CNN-BiLSTM-CRF model with opinion expression features ranks first in the results of this study.