Holder and Target Identification on Opinion Text Using Deep Neural Networks

EasyChair Preprint 9121

Abstract

The development of social media platforms has made it possible for everyone to be able to express their opinions online. Therefore, various techniques have been developed to extract the information in opinion texts. Opinion role labeling (ORL) aims to identify opinion holder and opinion target within documents. We propose the deep learning models to identify opinion holder and opinion target given opinion text. Based on the experiments using MPQA as training data, we report that the use of Convolutional Neural Network (CNN) architecture for character level feature extraction can increase the F1-score of the BERT-BiLSTM-CRF base as baseline model by 3%. In addition of an opinion expression feature on the model can significantly increase the F1-score of the baseline model by 20%.

Keyphrases: Opinion Target, deep learning, opinion holder, opinion role labeling

@booklet{EasyChair:9121,
  author    = {Mohamad Mirza Maulana Ikhsan and Fariska Zakhralativa Ruskanda},
  title     = {Holder and Target Identification on Opinion Text Using Deep Neural Networks},
  howpublished = {EasyChair Preprint 9121},
  year      = {EasyChair, 2022}}