Hint-Embedding Attention-Based LSTM for Aspect Identification Sentiment Analysis

Abstract

Aspect identification became an important task for aspect-based sentiment analysis. Previous approaches realized the importance of aspect identification in aspect-level sentiment analysis task. To this aim, there are different approaches proposed including rule-based and supervised learning based. Rule-based methods introduce rule mining based on features engineering, while supervised methods consider it as multi-task text classification problem. However, aspect identification is still a challenge from two perspectives: detecting the implicit aspect and mapping aspect-term into category. In this paper, we propose a novel neural network approach with Hint-embedding that aims at exploring the connection between an aspect and its semantic content in the sentence. Attention mechanism is designed to focus on different parts of a sentence based on aspects’ indicators. We experiment on benchmark datasets (SemEval 2014 task 4 restaurant and SemEval 2016 task 5 laptop), and results show that our model achieves considerable performance on aspect identification task.