Extracting Opinion Targets Using Attention-Based Neural Model

Abstract

Extracting opinion-target expression is a core subtask to perform aspect-based sentiment analysis which aims to identify the discussed aspects within a text associated with their opinion targets and classify the sentiment as positive, negative, or neutral. This paper proposes a deep learning model to tackle the opinion-target expression extraction task. The proposed model is composed of bidirectional long short-term memory as an encoder, long short-term memory as a decoder with an attention mechanism, and conditional random fields. This model, which operates at the sentence level, is designed to extract opinion targets for the Arabic language. The proposed model’s performance is evaluated using SemEval-2016 annotated dataset for the hotels’ domain. Experimental results show that the proposed model outperforms the baseline and the prior works, where it achieved an F1 measure of 72.83%.